29.04.03 · psychology / learning-memory

Memory systems: episodic, semantic, procedural; encoding, consolidation, retrieval

stub3 tiersLean: nonepending prereqs

Anchor (Master): Tulving, E. — Episodic and semantic memory (1972)

Intuition Beginner

Memory is not a single thing. It is several systems that work differently. Episodic memory records your personal experiences: what you did yesterday, your first day at school. Semantic memory stores general knowledge: the capital of France, the meaning of "photosynthesis." Procedural memory holds skills you perform automatically: riding a bike, typing on a keyboard.

Short-term memory holds a few items for just seconds; long-term memory can last a lifetime. Working memory, the modern version of short-term memory, actively manipulates information — like doing mental arithmetic or reordering a list in your head. How you encode information matters. Deeply processing material (thinking about its meaning) produces better memory than shallow processing (noticing only surface features). Sleep helps consolidate the memories you formed during the day.

Visual Beginner

Figure: The memory landscape in one view. The Atkinson-Shiffrin flow gates information from sensory to short-term to long-term storage; Squire's taxonomy separates declarative from nondeclarative systems; Craik-Lockhart levels of processing predict depth-driven retention; Baddeley's working memory model decomposes short-term holding into specialised subsystems; and consolidation unfolds on two timescales — synaptic over minutes-to-hours and systems over weeks-to-years.

Worked example Beginner

Three memory systems in action

Imagine you spend one afternoon learning to play a new chord on guitar from a video lesson. Three memory systems are working at once, and you can tell them apart by what each stores.

System	What it holds	Example from the lesson
Episodic	Personal events tied to time and place	Remembering sitting on your bed at 4pm watching the instructor
Semantic	General facts, divorced from when you learned them	Knowing a C-major chord is C, E, and G played together
Procedural	Motor and cognitive skills	Your fingers forming the chord shape without thought

The episodic detail fades fastest; the semantic fact may last for years; the procedural skill, once automatic, can survive decades of disuse. That is why you can forget exactly when you learned to ride a bike yet still ride one.

Deep versus shallow encoding

A classic experiment gives people a list of words and asks one of three questions per word. Later, recall is tested.

Question type	Depth	Example question	Recall (%)
Structural (shallow)	Low	"Is the word in capital letters?"	~20
Phonemic	Medium	"Does it rhyme with 'train'?"	~38
Semantic (deep)	High	"Does it fit the sentence 'The dog ___'?"	~65

Thinking about meaning leaves a stronger trace than thinking about appearance or sound. This is the levels of processing effect, and it is one of the most reliable findings in memory research ^{[source pending]}.

Check your understanding Beginner

Exercise (easy, multiple choice).

You learned the French word "chien" (dog) years ago in a classroom. Today you still know its meaning but cannot remember the lesson at all. Which memory system holds the meaning, and which held the now-faded lesson?

A. Semantic holds the meaning; episodic held the lesson. B. Episodic holds the meaning; procedural held the lesson. C. Procedural holds the meaning; semantic held the lesson. D. Semantic holds the meaning; procedural held the lesson.

Hint

One system stores general facts stripped of context; the other stores personally experienced events located in time and place.

Answer

Option A. The word's meaning is general knowledge independent of when you learned it, so it lives in semantic memory. The classroom lesson — a specific event at a specific time — was an episodic memory, and that detail has faded. Episodic detail typically decays faster than semantic knowledge, which is why you can keep a fact long after losing the context in which you learned it.

Formal definition Intermediate

The vocabulary of memory systems is standardised across the anchor texts ^{[source pending]}. These terms identify empirically dissociable processes, not mere labels.

The multi-store model (Atkinson and Shiffrin, 1968). Information flows through three stores, gated by control processes.

Store	Capacity	Duration	Gating process
Sensory register	Large (all modalities)	under 1 s	Attention selects a fraction
Short-term / working memory	Limited (Miller's $7 \pm 2$ ; Cowan's $\approx 4$ )	under 30 s without rehearsal	Rehearsal and elaboration
Long-term memory	Effectively unlimited	Potentially permanent	Encoding and consolidation

Capacity estimates have tightened over time: Miller's (1956) famous seven items assumed chunking; Cowan (2001) argued the true focus limit is nearer four chunks when rehearsal is controlled.

Long-term memory systems (Squire's taxonomy).

Declarative (explicit) — consciously recollected, requires the medial temporal lobe.
- Episodic — personally experienced events, located in time and place; supports mental time travel.
- Semantic — generic facts and meaning, stripped of episodic context.
Nondeclarative (implicit) — expressed through performance, not recollection; does not require the medial temporal lobe.
- Procedural — skills and habits (basal ganglia, cerebellum).
- Priming — prior exposure speeds later processing.
- Conditioning — learned CS–CR associations (amygdala, cerebellum).
- Nonassociative — habituation and sensitisation.

Encoding. The transformation of perceived information into a storable representation. The levels of processing framework (Craik and Lockhart, 1972) ranks structural (shallow) < phonemic < semantic (deep) processing in retention strength. The encoding specificity principle (Tulving and Thomson, 1973) states that retrieval succeeds in proportion to the overlap between cues present at encoding and cues available at retrieval. Transfer-appropriate processing refines this: memory improves when the type of processing at retrieval matches the type at encoding, regardless of absolute depth.

Consolidation. The stabilisation of an initially labile trace. Synaptic consolidation completes within minutes to hours and depends on protein synthesis. Systems consolidation unfolds over weeks to years as memories gradually become independent of the hippocampus. Reconsolidation occurs when a retrieved memory re-enters a labile state and must be re-stored, opening a window for modification.

Retrieval. Accessing stored information. The principal tests, in roughly ascending order of cue support: free recall (no cues), cued recall (a related cue is given), recognition (the target is presented, decide if it was studied), and savings (relearning is faster than original learning, the most sensitive measure). The tip-of-the-tongue state is a partial-retrieval failure: the trace is stored but the cue is insufficient to complete access.

Spacing effect. Distributed practice (study sessions separated by gaps) yields stronger long-term retention than massed practice (the same total time crammed into one session). The testing effect is the complementary finding that retrieving material during study improves later memory more than restudying it does, even without feedback.

Amnesias. Anterograde amnesia is the inability to form new long-term memories after the insult (classic in patient HM). Retrograde amnesia is the loss of memories formed before the insult, typically with a temporal gradient (recent memories lost, remote memories spared — Ribot's law).

Key model Intermediate

Baddeley and Hitch's (1974) working memory model replaced the passive short-term store with an active, multi-component system, and it remains the most influential account of how information is held and manipulated online ^{[source pending]}.

Core architecture. Working memory comprises a limited-capacity attentional controller (central executive) supervising two modality-specific slave systems and (added later) an integrative buffer:

Component	Function	Capacity limit	Neural substrate
Central executive	Attentional control, task switching, inhibition	Limited	Prefrontal cortex
Phonological loop	Holds verbal/acoustic info ~2 s via subvocal rehearsal	What can be said in ~2 s	Left perisylvian cortex
Visuospatial sketchpad	Holds visual and spatial information	Limited	Right parietal / occipital
Episodic buffer (added 2000)	Integrates codes from the slave systems and LTM into coherent episodes	Limited	Prefrontal / medial temporal

What the components explain. The phonological loop explains the word-length effect (more short words recalled than long ones, because each fits the two-second rehearsal window) and the articulatory suppression effect (forcing concurrent verbalisation fills the loop and abolishes the benefit of rehearsal). The visuospatial sketchpad explains why tracking a moving target disrupts mental imagery but not verbal memory, and vice versa. The episodic buffer resolves how we hold bound representations (a face with a name with a location) that no single slave system could store alone.

Working memory capacity (WMC). Individual differences in WMC, measured by complex span tasks (reading span, operation span), predict reading comprehension, reasoning (Raven's matrices), fluid intelligence, the ability to resist attentional capture, and the control of mind wandering. WMC is not intelligence, but it is among the strongest single psychometric predictors of higher cognition — a major bottleneck through which controlled thought must pass.

Relation to the multi-store model. Working memory does not abolish the Atkinson-Shiffrin stores so much as re-describe the short-term store as active and composite. Long-term memory remains a separate, vast system; the episodic buffer is the interface through which working memory and long-term memory exchange bound representations.

What the model does not capture. The central executive was long criticised as a homunculus — an ill-specified "does-everything" component. Later fractionations (Miyake and Friedman's three executive functions: shifting, updating, inhibition) decompose the executive into measurable, partly independent processes. The model also underspecifies the dynamics of encoding into long-term memory and the role of emotion and motivation, which the Master tier takes up through amygdalar modulation and tagging-and-capture accounts.

Exercises Intermediate

Exercise 1 (easy, multiple choice).

A participant holds a seven-digit phone number in mind. While rehearsing it, she is asked to tap keys in a rhythm. She repeats the digits correctly. Now she is instead asked to silently recite "the the the" continuously while holding the number. She now fails. Which working-memory component has been disrupted, and why?

A. Visuospatial sketchpad, because rhythmic tapping used it. B. Phonological loop, because subvocal rehearsal was blocked. C. Central executive, because dual-task load exhausted it. D. Episodic buffer, because binding failed.

Hint

Which component holds verbal material, and what process keeps it from decaying?

Answer

Option B. Continuously reciting "the the the" occupies the subvocal rehearsal route (articulatory suppression), so the digits in the phonological loop decay past the two-second window before they can be refreshed. Rhythmic tapping (a motor/spatial task) did not disrupt verbal rehearsal, which is why it was harmless. This dissociation is a classic demonstration that the phonological loop depends on inner speech.

Exercise 2 (easy, multiple choice).

A patient can form new procedural memories (he learns new motor skills normally) but cannot form new declarative memories (he cannot recall anyone he met yesterday). This dissociation most directly supports which claim?

A. Short-term and long-term memory are the same system. B. Declarative and nondeclarative memory depend on separable brain systems. C. Procedural memory is just very well-rehearsed semantic memory. D. Consolidation requires sleep.

Answer

Option B. The dissociation — preserved nondeclarative (procedural) learning alongside devastated declarative memory — is the defining feature of medial-temporal-lobe amnesia (patient HM is the archetype). It shows the two are supported by different brain systems: procedural memory by the basal ganglia and cerebellum, declarative memory by the hippocampus and surrounding medial temporal cortex. Were they one system, both would fail together.

Exercise 3 (medium, application).

A student has six hours to study for an exam and can choose between (a) one six-hour block the day before, or (b) six one-hour blocks spread across two weeks. Using the spacing effect and the testing effect, argue which schedule yields better long-term retention, and give one specific study technique the student should use within each session.

Hint

Distributed practice beats massed practice, and retrieval beats rereading.

Answer

Schedule (b), the spaced one-hour sessions, yields better long-term retention. The spacing effect is one of the most replicated findings in cognitive psychology: distributing study across time produces stronger memory than massing the same total time, because each re-encounter occurs after some forgetting, forcing effortful reconstruction that strengthens the trace and provides additional encoding contexts. Within each session the student should use retrieval practice (self-testing with flashcards or practice questions) rather than passive rereading: the testing effect shows that actively retrieving material produces more durable learning than restudying it, even without feedback. Combining spacing with testing is the single best-evidenced study strategy available.

Exercise 4 (medium, analysis).

A diver learns a list of words underwater and is later tested either underwater or on land. Recall is better underwater. Name the principle that predicts this result and explain, in 3–4 sentences, why the type of cue matters as much as the environment.

Answer

This is context-dependent memory, a direct consequence of Tulving's encoding specificity principle: retrieval succeeds in proportion to the overlap between cues present at encoding and cues available at retrieval. The underwater environment served as a retrieval cue encoded along with the words, so reinstating it at test re-activates the original trace. The deeper point of encoding specificity is that what counts as a "good" cue is defined relative to what was encoded — the type of processing (semantic, phonemic, contextual) at encoding determines which cues will be effective at retrieval. This is why transfer-appropriate processing matters: a cue helps only when it engages the same representational code that was laid down during learning.

Exercise 5 (hard, critical evaluation).

A clinician tells a patient with mild anterograde amnesia that because she cannot consciously recall new events, she "cannot learn anything new at all." Drawing on the declarative/nondeclarative distinction and one specific experimental result, evaluate this claim in 4–6 sentences.

Hint

If the claim were true, what should happen when an amnesic patient is taught a motor skill?

Answer

The claim is false. Anterograde amnesia specifically impairs declarative (episodic and semantic) memory — the ability to consciously recollect facts and events — while nondeclarative memory is often preserved. Patient HM, the canonical case, could not recognise the clinicians he saw daily, yet he learned mirror-drawing and other motor skills at a normal rate, improving session over session while disclaiming any memory of ever having done the task. This dissociation shows that learning and conscious recollection are separable: the patient acquires the skill (procedural memory, basal ganglia/cerebellum) without recollecting the learning episodes (declarative memory, hippocampus). The clinician's error is to equate "memory" with its declarative form alone, ignoring the nondeclarative systems that continue to function.

Advanced results Master

Squire's taxonomy and the medial temporal lobe memory system

Squire's (1992) division of long-term memory into declarative (episodic + semantic) and nondeclarative (procedural, priming, conditioning, nonassociative) systems is grounded in double dissociations of the kind lesion studies supply ^{[source pending]}. Declarative memory depends on a medial temporal lobe memory system: the hippocampus proper together with the adjacent entorhinal, perirhinal, and parahippocampal cortices. Damage confined to this system abolishes the ability to form new declarative memories while leaving nondeclarative learning intact. The components are not interchangeable: perirhinal cortex is disproportionately involved in object familiarity (perceptual "knowing"), parahippocampal cortex in scene and context processing, and the hippocampus proper in associating items with their spatiotemporal context ("remembering").

Patient HM and Patient KC

The empirical keystone is Patient HM (Henry Molaison). Scoville and Milner (1957) reported that a bilateral medial temporal lobectomy, performed to relieve intractable seizures, left HM with devastating anterograde amnesia: he could not form new declarative memories, could not consciously recall events from roughly a decade before the surgery (temporally graded retrograde amnesia), yet retained his pre-surgical remote memories, his IQ, his personality, and crucially his ability to learn new procedural skills — improving at mirror-drawing over days while insisting he had never seen the task before ^{[source pending]}. HM established that the medial temporal lobes are necessary for declarative but not procedural memory, and that memory is not unitary.

Patient KC (Kent Cochrane) furnished the complementary dissociation within declarative memory. Following a motorcycle accident that produced widespread medial temporal damage, KC lost episodic memory almost entirely — he could not mentally re-experience any personal past event — yet retained substantial semantic knowledge: he knew facts about the world, including facts about his own life, but stripped of any first-person temporal context ^{[source pending]}. KC's case supports Tulving's proposal that episodic and semantic memory are functionally and neurally separable: episodic memory requires hippocampally-mediated binding of an event to its spatiotemporal context, while semantic memory can be supported by neocortical representations that survive hippocampal loss.

Pattern separation and pattern completion

Within the hippocampus, two computational operations are distinguished. Pattern separation, attributed to the dentate gyrus, transforms overlapping inputs into dissimilar representations so that similar experiences (parking in the same lot on two different days) are stored as distinct memories and do not interfere. Pattern completion, attributed to CA3 with its extensive recurrent collaterals, allows a partial cue (the smell of a grandmother's kitchen) to reinstate the full stored representation. The balance between these operations determines whether an experience is stored as novel (separation) or assimilated to an existing trace (completion), and failures of either produce specific pathologies — from over-completion causing false recognitions, to impaired separation in ageing and anxiety.

Sharp-wave ripples, replay, and systems consolidation

During slow-wave sleep and quiet wakefulness, the hippocampus generates sharp-wave ripples: brief, high-frequency population bursts accompanied by the sequential reactivation of neuronal ensembles that fired together during prior experience. This replay is the leading neural candidate for the mechanism of systems consolidation: by reactivating episodic traces offline, the hippocampus is thought to drive their gradual redistribution to neocortical networks, until the memory can be retrieved without hippocampal involvement. Disrupting ripples selectively impairs spatial and episodic learning, and the content of replay can even be biased toward future trajectories (preplay), linking this mechanism to planning.

The systems transformation hypothesis extends this picture: as episodic memories consolidate over weeks to years, they do not merely move from hippocampus to neocortex, they are transformed. The hippocampally-bound, context-rich episodic trace loses its detail and its spatiotemporal specificity, and what remains in neocortex is the semantic gist — the general knowledge extracted from many episodes. The same mechanism that consolidates also schematises, which is why old autobiographical memories come to feel like facts about your life rather than re-lived experiences.

Reconsolidation and memory malleability

The classical view treated consolidation as a one-time process: once a trace was stabilised, it was fixed. Nader, Schafe, and LeDoux (2000) overturned this for fear memory by showing that retrieving a consolidated memory renders it labile again, opening a protein-synthesis-dependent window during which the trace can be weakened, strengthened, or modified before being reconsolidated ^{[source pending]}. Infusing a protein-synthesis inhibitor into the amygdala immediately after retrieval (but not without retrieval) erased the behavioural expression of a previously consolidated fear memory. Reconsolidation is now a general mechanism: every act of remembering is also, potentially, an act of re-writing. This supplies a neural substrate for the reconstructive malleability that Loftus documented at the behavioural level, and it reframes therapeutic exposure not as erasure but as the laying down of a modified trace.

Engram cells

The physical substrate of a memory — the engram — had been hypothesised since Lashley's (failed) search for it. Tonegawa and colleagues (2012) provided the first convincing demonstration by labelling, in the dentate gyrus, the cells that were active during fear conditioning in a context-specific manner, then optogenetically reactivating that labelled ensemble in a different, neutral context ^{[source pending]}. The animal froze — it "remembered" the shock — even though no external cue was present, proving that the labelled ensemble is both sufficient (its artificial activation evokes the memory) and necessary (silencing it abolishes recall). Subsequent work showed that ensembles can be reactivated to induce false memories (activating a context-A engram while shocking in context B produces a fear of A), that engram cells exist in silent forms before consolidation, and that the connectivity between engram ensembles across brain regions (hippocampus, amygdala, prefrontal cortex) constitutes the physical memory trace.

Synaptic tagging and capture

Frey and Morris (1997) proposed the synaptic tagging and capture hypothesis to explain how weakly stimulated synapses, which alone would not produce long-term potentiation, can nonetheless be stabilised if strong stimulation occurs elsewhere in the same neuron within a critical time window. The weak stimulation sets a local "tag"; the strong stimulation triggers the synthesis of plasticity-related proteins that are captured by the tagged synapses, stabilising them. This reconciles two facts: that long-term memory requires protein synthesis, and that memories can be formed from brief individual experiences. It also provides a mechanism by which the emotional salience of one event (driving broad protein synthesis) can enhance the consolidation of unrelated but temporally coincident material — a cellular analogue of state-dependent memory enhancement.

Emotional memory and amygdalar modulation

The amygdala does not store memories but modulates their consolidation. Emotionally arousing events are better remembered than neutral ones, and the effect is mediated by amygdala-driven hormonal modulation (adrenaline, cortisol) of hippocampal consolidation. The relation is non-monotonic, following the Yerkes-Dodson law: moderate arousal optimises memory, while very high arousal (trauma) can impair hippocampal encoding even as it strengthens amygdala-dependent fear learning. This asymmetry explains why a traumatic event may be vividly remembered as a fragmentary emotional image yet poorly encoded in its sequential, contextual detail — a pattern central to PTSD.

False memory implantation

Loftus extended the misinformation paradigm to its limit: under suggestive conditions, participants can be led to develop vivid, confident memories of events that never occurred, including childhood experiences ("Lost in the Mall") and impossible events. Implantation succeeds more often when the suggested event is plausible, when an authority figure (a trusted relative, a therapist) corroborates it, and when imagination is recruited to elaborate the false detail. The phenomenology of an implanted false memory — its vividness, its emotional charge, the subject's confidence — is indistinguishable from that of a true memory, which is why the subject cannot self-diagnose the error. The convergence with reconsolidation research is direct: both lines show that remembering is constructive, and that what is constructed is felt as given.

Limits the framework imposes

Memory-systems research is one of the most mechanistically developed areas of psychology, with behavioural dissociations, lesion data, single-unit recording, optogenetics, and molecular biology all converging on a multi-system architecture. The price is that the framework is, at present, a patchwork: the working-memory model, the medial-temporal declarative system, procedural learning, and reconsolidation are each supported but are not yet unified into a single quantitative theory the way Rescorla-Wagner unified conditioning. The nontrivial empirical claim is that memory is not a faculty but a federation, and that each member of the federation has its own neural substrate, its own time course, and its own characteristic pathologies.

Connections Master

Learning and memory 29.04.01 is the immediate prerequisite. This unit deepens its memory-systems strand (encoding, storage, retrieval) into a multi-system architecture; the conditioning strand is taken up in 29.04.02 pending. The companion unit 29.04.04 pending (pending) extends this material to forgetting and false memory in depth, to which this unit is the systems foundation.
Conditioning 29.04.02 pending connects through nondeclarative memory: conditioned CS–CR associations are themselves an implicit memory system supported by the amygdala and cerebellum. The prediction-error machinery of Rescorla-Wagner and the dopamine RPE described there run largely outside the declarative system described here, which is why conditioning can survive medial-temporal damage that abolishes episodic memory.
Neuroscience [29.02.NN] (pending) supplies the substrate. The hippocampal formation (dentate gyrus, CA3, CA1), the medial temporal cortices, the basal ganglia, and the cerebellum realise the systems catalogued here; long-term potentiation is the leading synaptic-plasticity candidate; and sharp-wave ripples and dopaminergic modulation are the mechanisms of consolidation and emotional tagging respectively.
Cognition and intelligence [29.05.NN] (pending) inherits working-memory capacity as a principal predictor of fluid intelligence and reasoning. Baddeley's central executive and Miyake's executive-function fractionation are central constructs in the attention-and-control literature.
Development [29.06.NN] (pending) connects through the lifespan trajectory of memory systems: infantile and childhood amnesia (immature hippocampus), the development of working memory through adolescence (maturing prefrontal cortex), and semanticisation of episodic memory in ageing.
Psychological disorders [29.09.NN] (pending) inherits these systems directly: PTSD as a failure of fear-memory extinction and reconsolidation, Alzheimer's disease as a progressive declarative-memory disorder beginning in the entorhinal cortex, and dissociative amnesia as a disorder of autobiographical retrieval.
Therapy and treatment [29.10.NN] (pending) applies the consolidation and reconsolidation machinery: exposure therapy as reconsolidation-updating of fear traces, and emerging pharmacological approaches (propranolol during retrieval) as direct interventions on the reconsolidation window.
Statistics and learning theory [26.NN, 25.NN] (pending) connect at the computational level: pattern separation and completion are formalisable as sparse-coding and attractor-network operations, and modern machine-learning models of episodic memory (hippocampal-inspired associative memories) draw on the same mathematics.

Historical & philosophical context Master

William James (1890) drew the original distinction between primary memory (what is held in consciousness now) and secondary memory (what must be retrieved from the past) — a forerunner of the short-term/long-term division that Atkinson and Shiffrin (1968) would formalise into the multi-store model that dominated the field for two decades ^{[source pending]}. The multi-store model's strength was its testable architecture (separable stores with characteristic capacities and durations); its weakness was that it treated the short-term store as passive, which the data on mental arithmetic, comprehension, and reasoning would not allow.

Baddeley and Hitch (1974) introduced working memory to fix exactly this. By showing that a concurrent memory load interfered with reasoning and comprehension only partially — not catastrophically, as a single passive store predicted — they argued for a multi-component system with a flexible executive. The phonological loop, visuospatial sketchpad, and central executive were the original components; the episodic buffer was added in 2000 to handle the binding problem the original model could not solve. The model's longevity reflects its combination of falsifiable architecture with broad empirical coverage.

Tulving's (1972) distinction between episodic and semantic memory arose from a different line of evidence: the realisation that "remembering" a personal event and "knowing" a fact are phenomenologically and functionally distinct. Tulving later tied episodic memory to autonoetic consciousness — the capacity for mental time travel — and argued that episodic memory is a late-evolving, evolutionarily advanced system that builds on the older semantic system. Patient KC's dissociation supplied the critical lesion evidence.

The lesion tradition that produced Squire's taxonomy rests on a handful of patients whose tragedies became foundational data. Patient HM's surgery was performed before informed consent took its modern form; he could not consent meaningfully to an experimental procedure, and for decades he was studied as "HM," his identity shielded until after his death in 2008, when it was revealed that he was Henry Molaison. His contribution to neuroscience is immeasurable; the ethical questions about how he was studied are permanent. The same is true of Patient KC and of the reconsolidation studies that depended on animal fear conditioning.

The modern engram program — Tonegawa's optogenetic reactivation of memory ensembles — realises a programme Richard Semon named and Lashley sought in vain for sixty years. The technical breakthroughs (immediate-early-gene labelling, channelrhodopsin targeting) made it possible to do what Lashley could not: identify, manipulate, and reactivate the specific cells that constitute a memory. The result is a convergence of molecular biology, systems neuroscience, and behaviour on a question psychology had posed since the 19th century: where, physically, is a memory?

The reconsolidation finding (Nader, Schafe, and LeDoux, 2000) reopened a question the classical consolidation theory had declared settled. If every retrieval re-opens the trace for modification, then memory is never literally "stored" in the sense a recording is stored — it is continually reconstructed, and the boundary between remembering and re-learning is blurred. This convergence with Loftus's behavioural work on reconstructive memory and false-memory implantation is one of the deepest results in the field: the constructive nature of memory is not a failure mode but its normal mode of operation, with consequences for law (eyewitness testimony), therapy (recovered and implanted memories), and education (testing as learning rather than merely assessment).

Bibliography Master

Atkinson, R. C. & Shiffrin, R. M. — "Human memory: a proposed system and its control processes", in Spence & Spence (eds.), The Psychology of Learning and Motivation, Vol. 2 (Academic Press, 1968), pp. 89–195.
Baddeley, A. D. & Hitch, G. — "Working memory", in Bower (ed.), The Psychology of Learning and Motivation, Vol. 8 (Academic Press, 1974), pp. 47–89.
Baddeley, A. — "The episodic buffer: a new component of working memory?", Trends in Cognitive Sciences 4, 417–423 (2000).
Craik, F. I. M. & Lockhart, R. S. — "Levels of processing: a framework for memory research", Journal of Verbal Learning and Verbal Behavior 11, 671–684 (1972).
Tulving, E. — "Episodic and semantic memory", in Tulving & Donaldson (eds.), Organization of Memory (Academic Press, 1972), pp. 381–403.
Tulving, E. & Thomson, D. M. — "Encoding specificity and retrieval processes in episodic memory", Psychological Review 80, 352–373 (1973).
Squire, L. R. — "Declarative and nondeclarative memory: multiple brain systems supporting learning and memory", Journal of Cognitive Neuroscience 4, 232–243 (1992).
Scoville, W. B. & Milner, B. — "Loss of recent memory after bilateral hippocampal lesions", Journal of Neurology, Neurosurgery, and Psychiatry 20, 11–21 (1957).
Tulving, E., Hayman, C. A. G. & Macdonald, C. A. — "Long-lasting perceptual priming and semantic learning in amnesia: a case experiment", Journal of Experimental Psychology: Learning, Memory, and Cognition 17, 595–617 (1991).
Miller, G. A. — "The magical number seven, plus or minus two: some limits on our capacity for processing information", Psychological Review 63, 81–97 (1956).
Cowan, N. — "The magical number 4 in short-term memory: a reconsideration of mental storage capacity", Behavioral and Brain Sciences 24, 87–114 (2001).
Nadel, L. & Moscovitch, M. — "Memory consolidation, retrograde amnesia and the hippocampal complex", Current Opinion in Neurobiology 7, 217–227 (1997).
Nader, K., Schafe, G. E. & LeDoux, J. E. — "Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval", Nature 406, 722–726 (2000).
Liu, X., Ramirez, S., Pang, P. T., Puryear, C. B., Govindarajan, A., Deisseroth, K. & Tonegawa, S. — "Optogenetic stimulation of a hippocampal engram activates fear memory recall", Nature 484, 381–385 (2012).
Frey, U. & Morris, R. G. M. — "Synaptic tagging and long-term potentiation", Nature 385, 533–536 (1997).
Buzsáki, G. — "Hippocampal sharp wave-ripple: a cognitive biomarker for episodic memory and planning", Hippocampus 25, 1073–1188 (2015).
Yerkes, R. M. & Dodson, J. D. — "The relation of strength of stimulus to rapidity of habit-formation", Journal of Comparative Neurology and Psychology 18, 459–482 (1908).
Roediger, H. L. & Karpicke, J. D. — "Test-enhanced learning: taking memory tests improves long-term retention", Psychological Science 17, 249–255 (2006).
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T. & Rohrer, D. — "Distributed practice in verbal recall tasks: a review and quantitative synthesis", Psychological Bulletin 132, 354–380 (2006).
Loftus, E. F. & Pickrell, J. E. — "The formation of false memories", Psychiatric Annals 25, 720–725 (1995).
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A. & Wager, T. D. — "The unity and diversity of executive functions and their contributions to complex 'frontal lobe' tasks: a latent variable analysis", Cognitive Psychology 41, 49–100 (2000).
Myers, D. G. & DeWall, C. N. — Psychology, 13th ed. (Worth, 2021), Ch. 8.
Gleitman, H., Gross, J. & Reisberg, D. — Psychology, 8th ed. (Norton, 2011), Ch. 7.