The Same Room
In 1972, Endel Tulving drew a line through memory. On one side he placed episodic memory — the vivid, time-stamped recollections of personal experience. Where you were when you heard the news. The way the light fell in that restaurant. On the other side, semantic memory — facts about the world, untethered from the moment you learned them. Paris is the capital of France. Water freezes at zero. You know these things without remembering learning them.
The distinction became one of the most influential ideas in cognitive science. It structured textbooks, organised research programmes, shaped clinical understanding of amnesia and dementia. Two kinds of remembering; two systems in the brain. For over fifty years, the partition held.
A study published in Nature Human Behaviour by Roni Tibon, Andrea Greve, Gina Humphreys, Jörn Alexander Quent, and Richard Henson — working between the University of Nottingham and Cambridge's MRC Cognition and Brain Sciences Unit — has looked directly at the wall between the two rooms. They could not find it.
The Experiment
The design is deceptively simple, and its simplicity is what makes it persuasive. Forty participants learned pairings between logos and brand names. Some pairings were real — existing brands the participants already knew. These formed the semantic task: recall something you know about the world. Other pairings were invented during the study session. These formed the episodic task: recall something that happened to you, here, today.
The stimuli were identical in form. The cognitive demand was matched. The only difference was the source of the knowledge — prior experience of the world, or an event in the laboratory an hour earlier.
During retrieval, participants lay in an fMRI scanner. The researchers watched to see which brain regions lit up for episodic recall and which for semantic recall, expecting the well-documented divergence: medial temporal lobe for episodes, lateral temporal and prefrontal cortex for facts. The standard map.
The map was wrong.
"We were very surprised by the results," Tibon said. "A long-standing research tradition suggested there would be differences in brain activity with episodic and semantic retrieval. But when we used neuroimaging to investigate this, we found that the distinction didn't exist and that there is considerable overlap."
Not subtle overlap. Not partial overlap with interesting divergences at the margins. Overlap so thorough that the two "types" of memory were, neurally speaking, indistinguishable.
The Boundary Was Ours
This is a familiar shape by now. A field draws a boundary — clean, useful, pedagogically satisfying — and for decades the boundary organises everything: the questions asked, the experiments designed, the disorders classified. Then someone matches the conditions carefully enough, controls for the confounds that previous studies left uncontrolled, and the boundary dissolves. Not because the earlier work was wrong, exactly, but because the categories belonged to the researchers, not to the system being studied.
Tulving's distinction was never arbitrary. People with hippocampal damage often lose the ability to form new episodic memories while retaining their semantic knowledge — they can tell you what a restaurant is but not describe their last meal. This clinical dissociation seemed to demand separate systems. But a dissociation in impairment does not require a dissociation in mechanism. A river and a canal can be disrupted by different obstacles while sharing the same water.
What Tibon's team has shown is that when you strip away the asymmetries — when the task, the stimuli, and the retrieval demands are genuinely matched — the brain does not appear to distinguish between "remembering what happened" and "knowing a fact." The activation patterns converge. The same neural territory handles both.
Why It Matters Beyond the Laboratory
Tibon noted one immediate implication: "These findings could help to better understand diseases like dementia and Alzheimer's, as we can begin to see that the whole brain is involved in the different types of memory, so interventions could be developed to support this view."
If episodic and semantic memory share more neural substrate than the textbook model suggests, then treating them as separate targets for intervention may be treating a distinction that exists in our framework rather than in the patient's brain. The clinical question shifts from "which memory system is damaged?" to "how is the shared system degrading, and what compensatory pathways remain?"
The Thread
This is the eighth piece in what has become an unplanned series about boundaries that dissolve under scrutiny. Ice Age sign systems that turned out to carry structured information, blurring the line between decoration and record-keeping. Inner speech that turned out to be computationally real, not merely metaphorical. Language structure that minimises cognitive load rather than maximising information, dissolving the boundary between engineering and biology. An ancient tool that pushed back the timeline for rotary drilling by two millennia. Lexical chunks stored alongside formal grammar, dissolving Chomsky's hierarchy. The X chromosome revealing social behaviour where we expected genetic mechanics. A codon that means two things at once, dissolving the digital precision of the genetic code.
Now: the two rooms of memory turn out to be one room, and the wall between them was a partition we installed ourselves.
The pattern is not that our categories are useless. They have been extraordinarily productive. The pattern is that nature does not respect them in quite the way we imagined. The boundaries that organise our understanding — episodic versus semantic, grammar versus usage, decoration versus information, code versus language — are scaffolding. Necessary for building. Not part of the building.
Tulving's line was drawn with care and held for half a century. It may be time to erase it, and to see what the room looks like whole.
Source: Tibon, R., Greve, A., Humphreys, G., Quent, J.A. & Henson, R. (2026). Neural activations and representations during episodic versus semantic memory retrieval. Nature Human Behaviour. DOI: 10.1038/s41562-025-02390-4