Types of Memory

Sensory memory

Before information can be remembered in any useful sense, it must survive long enough to be selected. Sensory memory is that survival window: an extremely brief, high-capacity buffer that holds a raw, unprocessed copy of what the senses have just delivered, giving attention a moment to pick out what matters.

The visual buffer is called iconic memory. Its properties were established by George Sperling in 1960 with an elegantly simple experiment. He flashed a grid of twelve letters for a fraction of a second and asked participants to report what they had seen. They could typically name only four or five, which had long been taken to mean that only four or five had registered. Sperling then changed the method: instead of asking for the whole grid, he sounded a tone immediately after the display to indicate which single row to report. Now participants could report almost any row accurately, which means all the letters had been available. What limited them was not perception but decay: by the time they had reported the first few, the rest had faded. Delaying the cue by even a third of a second destroyed the advantage. Iconic memory holds a lot, and holds it for less than a second.

The auditory equivalent, echoic memory, lasts longer, on the order of two to four seconds. This is why you can be asked a question while distracted, say "what?", and then answer it before the other person repeats themselves: the sound was still sitting in the buffer, waiting to be attended to.

Short-term and working memory

Information selected out of the sensory buffer enters a limited store that holds it for seconds. In the older literature this is called short-term memory, and it is essentially passive: a phone number held just long enough to dial it. Without rehearsal, the contents fade in around fifteen to thirty seconds.

Modern work has largely replaced the passive store with the richer notion of working memory, most influentially in the model of Alan Baddeley and Graham Hitch. Working memory is not a box that information sits in. It is an active system that holds information and does something with it: reversing a sequence of digits, keeping track of where you are in a multi-step instruction, holding the start of a long sentence while parsing its end. It is the mental workspace, and it depends heavily on the prefrontal cortex working with posterior sensory areas.

The capacity question: seven or four?

Nearly everyone has heard that short-term memory holds "seven plus or minus two" items. The figure comes from George Miller's 1956 paper, one of the most cited and most misread articles in psychology.

Miller's argument was not that people can hold seven things. It was about chunks: meaningful units, whose contents can be arbitrarily rich. The string 1, 9, 4, 5, 1, 9, 8, 9 is eight digits, but for someone who recognises 1945 and 1989 as dates it is two chunks. Miller's own title, "The magical number seven, plus or minus two", was partly ironic; he was struck by a recurring number across quite different tasks and was frank that he did not know why it recurred. The lesson of his paper is really that chunking multiplies effective capacity, not that seven is a hard ceiling.

When later researchers stripped out chunking and blocked rehearsal, the true limit turned out to be considerably smaller. In a systematic 2001 review, Nelson Cowan concluded that the capacity of the focus of attention in working memory is closer to four chunks in adults, and that the classic figure of seven reflected tasks in which participants could rehearse and group material. Four is the number to teach.

Long-term memory: the two branches

Long-term memory is where the taxonomy really matters. The primary split, drawn from the amnesia evidence and formalised by Larry Squire and colleagues, is between memory you can consciously bring to mind and declare, and memory that shows up only in what you can do.

The two branches differ not only in anatomy but in character. Declarative memory is acquired quickly, often in a single exposure, and it is flexible: what you learn in one context can be applied in another. Non-declarative memory is typically acquired gradually over many repetitions, and it is rigid: it tends to be expressed only in circumstances close to those in which it was learned.

Explicit memory: episodic and semantic

The division within declarative memory was proposed by Endel Tulving in 1972, and it has held up remarkably well.

Episodic memory is memory for events you personally experienced, bound to a time and a place: what you ate this morning, where you were when you heard some piece of news, the argument you had last Tuesday. Retrieving an episodic memory involves a kind of mental time travel, a re-experiencing of the event from your own perspective. Tulving called the accompanying state autonoetic consciousness, the sense of the self travelling through subjective time.

Semantic memory is memory for facts and concepts, held free of any record of where you acquired them. You know that Paris is the capital of France, that a chair has legs, that water freezes at zero degrees Celsius. You almost certainly have no memory of learning any of these. The knowledge has been stripped of its episodic origin.

The two interact constantly. Semantic knowledge is thought to be built up partly by abstracting across many episodes: repeated experiences of dogs eventually yield the concept "dog", independent of any particular dog. And episodic retrieval leans on semantic knowledge to reconstruct plausible detail, which is one route by which memories drift toward the typical and away from the true.

Implicit memory: skills, priming, conditioning

Non-declarative memory is not one system either. It is a collection of learning mechanisms that share only the negative property of not requiring conscious recollection.

The dissociation: what the amnesia evidence proves

A taxonomy is only as good as the evidence that the categories are really separate. Here the evidence is unusually strong, and it comes from patients.

The patient H.M., following bilateral medial temporal lobe surgery in 1953 and described by Scoville and Milner in 1957, could form essentially no new declarative memories. He could not learn new facts, could not recall events from moments earlier, and did not recognise the researchers who had worked with him for years. If declarative and non-declarative memory were the same system, his learning should have stopped entirely.

It did not. Brenda Milner gave him a mirror-drawing task: trace a shape while seeing only its mirror image, which initially produces clumsy, error-strewn attempts. Over successive days of practice his performance improved steadily and substantially, following the normal learning curve. And every single day he insisted he had never done the task before. He had no memory of practising, and yet the practice had unmistakably taken hold.

H.M. on his own, however, proves less than it first appears. A sceptic has an obvious escape route: perhaps the two tasks were not equally hard. Perhaps remembering an episode is simply more demanding than acquiring a motor skill, so any brain injury would damage the harder one first, and nothing about separate systems follows. To close that escape you need the mirror image: a patient who shows the opposite pattern.

That patient exists. In Huntington's disease and in Parkinson's disease the basal ganglia degenerate, and the result is precisely the reverse of H.M. Given a task in which an association has to be built up slowly across many trials of feedback, these patients fail to acquire it, while remembering the training sessions perfectly well and being able to describe what they were asked to do. Amnesic patients, given the same task, show the exact opposite: they learn the association at a normal rate while denying they have ever seen the task before. Two patient groups, two tasks, and the impairments run in opposite directions.

Prospective memory

One category sits awkwardly in the classic taxonomy but matters enormously in daily life: prospective memory, remembering to do something in the future. Taking a tablet at eight o'clock, passing on a message, posting a letter on the way past the box.

It is not simply declarative memory pointed forwards. Prospective memory has two components: the retrospective component, which is the content of the intention, and the prospective component, which is remembering to retrieve the intention at the right moment, in the absence of anyone asking you to. That second component depends on executive function and prefrontal circuits, and it is why prospective memory failures are so common in people who are busy, tired, or distracted, and so prominent after frontal-lobe damage. Prospective memory tasks divide further into event-based (do X when you see Y) and time-based (do X at four o'clock), and the time-based variety is notably harder, because nothing in the environment prompts you.

Myths worth discarding

Related topics