Skip to content

← Psychology

Memory

Chapter 9 · Learning and remembering

In this chapter 5 sections
  1. Three kinds of holding
  2. How much the mind can hold at once
  3. Getting it in
  4. Kinds of long-term memory
  5. Getting it out

A student at Carnegie Mellon University, in the United States, once agreed to come to a laboratory and do something rather dull. A psychologist would read him a string of random digits, one a second, and he would say them back in order. Then another string, and another, an hour a day, two to five days a week. The research papers call him SF.

On his first day SF could manage about seven digits, which is what most people manage. That is worth trying for yourself before reading on.

Try it

Hold the digits

Digits appear one at a time. When they stop, type them back in the same order. Each list you get right is one digit longer.

The digits will appear here
Digits in this list
—
Longest list right
—
The last list was
—

Press Start. Digits appear one at a time, one a second. When they stop, type them in the same order.

Whatever you scored, it says little about you: one run on one day swings with luck, tiredness and how hard you were looking, and a screen isn’t a voice reading aloud. But most adults who take the spoken test land somewhere around seven. SF did, at first.

Then he kept going. After more than 230 hours of practice, spread over nearly two years, he could hear 79 random digits once and say them all back without a mistake. His memory hadn’t grown. He had found a trick, and the trick explains most of this chapter. When the psychologists gave him letters instead of digits, his span was ordinary again.

To see what the trick was, it helps first to see how memory is put together.

Three kinds of holding

Memory is the set of ways the mind takes in information, keeps it, and gets it back when it is needed. Psychologists give the three steps names. Getting something in is encoding. Keeping it is storage. Getting it out again is retrieval. Forgetting a name at a wedding can be a failure at any of the three: you never really took it in, it didn’t last, or it is in there and won’t come out. A few things hold for memory everywhere. It is a set of processes, not a place. It takes in only what is attended to. It is narrow in the middle, where things are held for a moment, and roomy at the far end. It is organized by meaning, each thing linked to others. And it works through cues: a memory is found by way of something tied to it.

This way of describing the mind, as something that takes in information, works on it and puts out an answer, is called the cognitive or information-processing approach. It grew up when psychologists tired of an older rule, that a science of behavior may talk only about the stimulus going in and the response coming out. The cognitive psychologists put the organism back in the middle, so that what happens is stimulus, then the working of a mind, then response. Nepal’s Grade 11 course calls this the S-O-R model: stimulus, organism, response. Their favorite comparison was the computer, and it helped: it gave them models that make exact predictions, which experiments can test. It also misleads if pushed too far. A brain does a great many things at once, not one step after another, a computer has no feelings or motives to bend what it stores, and a list of digits in a laboratory is a long way from remembering your grandmother’s kitchen.

The model most books start from was set out in 1968 by two psychologists, Richard Atkinson and Richard Shiffrin. It has three stages: sensory memory, short-term memory and long-term memory.

Sensory memoryAlmost everything the senses take inAbout a second for sightShort-term memoryWhat you are thinking about nowAbout four chunks, 15–30 secondsLong-term memoryEverything you know and have doneNo known limit, for yearsattentionthe rest fadesencodingretrievalrehearsal
Figure 9.1 The three stages of memory. Attention passes a little of what the senses take in to short-term memory, which keeps it for seconds unless it is rehearsed. Encoding stores it in long-term memory, and retrieval brings it back.

Sensory memory is the briefest. Each sense holds on to what it has just taken in, nearly all of it, for a moment. In 1960 the psychologist George Sperling flashed a card of twelve letters and numbers, three rows of four, for a twentieth of a second. Asked to name them all, people could give only about four. But when a tone right after the flash told them which row to name, high, middle or low, they could give most of any row. Since they didn’t know which row would be asked for, nearly the whole card must have been there for a moment, fading before they could read it all out. Within about a second the advantage was gone. Hearing has a brief store of its own, which is how you can catch the last few words of a question you weren’t listening to.

Out of that flood, attention picks a little, and what it picks goes into short-term memory. This is what you are thinking about right now: the number you are about to dial, the start of this sentence. It is small, and it doesn’t last. Unless you keep repeating it, what is held there is mostly gone in 15 to 30 seconds, and in one classic experiment people asked to count backward after seeing three letters remembered them about 80% of the time after three seconds but only about 10% after eighteen. Why it goes, by fading or by being pushed out, is the next chapter’s question.

Long-term memory is the rest: every word you know, your way home, the day your sister was born. As far as anyone has found, it has no limit, and it can last a lifetime.

A good test of the model is to read someone a list of words and ask for as many as they can, in any order. People recall the first few words and the last few better than the middle. In the stage model this has an easy reading. The first words were repeated in the head more than the others, so they reached long-term memory, while the last ones are still in short-term memory when the list stops. If the person must count aloud for half a minute before recalling, the advantage of the last words disappears, as short-term memory empties. The pattern is called the serial position effect, primacy for the start and recency for the end, though not every detail of it fits the model.

How much the mind can hold at once

In 1956 the psychologist George Miller opened a famous paper with the line “My problem is that I have been persecuted by an integer.” The integer was seven. Again and again, in tests of how much people could hold in mind at once, it came out at about seven, give or take two. But Miller’s real point was what was being counted. It wasn’t digits or letters or bits of information. It was chunks: units that mean something to you. A beginner at Morse code hears each dot and dash separately. An expert hears whole letters, then whole words. The number of chunks stays about the same, and what grows is how much each chunk holds.

Later work brought the number down. When people are stopped from repeating things to themselves or from grouping them into bigger chunks, most can hold only about four chunks at once, somewhere from three to five. The seven in a digit test comes from the extra help of saying the digits over in your head and quietly grouping them.

Grouping is something everybody does without being taught. A Nepali mobile number is ten digits, more than most people can hold. Nobody holds it as ten. You hear 98, then a group of four, then the rest, and if the middle group is your uncle’s year of birth, it is one chunk and not four.

That was SF’s trick, on a grand scale. He was a runner, and he heard runs of digits as race times. Four digits such as 3492 became 3 minutes 49.2 seconds, a time near the world record for the mile. Later he added ages, so 896 was 89.6, a very old man. He never held more than a few digits at once in the ordinary way. Each group became something he already knew, stored in long-term memory, and over many months he built a fixed plan of where each group went, so he could find the groups again in order. His short-term memory hadn’t grown at all, which is why letters undid him: he had no race times for letters. A second runner, taught SF’s method, went further still and could recall 104 digits after more than three years of near-daily practice.

Short-term memory is now usually thought of as more than a shelf. In 1974 the psychologist Alan Baddeley and a colleague called it working memory, a workbench where information is held and also worked on, and they split it into parts. An inner voice, the phonological loop, holds a few words by repeating them. An inner eye, the visuospatial sketchpad, holds shapes and places. A central executive, in charge, decides what to attend to and switches between tasks. In 2000 Baddeley added a fourth part, the episodic buffer, which binds these together with what long-term memory supplies.

The evidence for the parts comes from doing two things at once. Two tasks of the same kind, like holding a list of words while reading, get in each other’s way. A task for the inner voice and one for the inner eye, like holding words while tracing a shape, barely interfere. Say something meaningless out loud over and over, “the, the, the”, and your memory for a list of words drops, because the inner voice is busy. The inner voice also seems to hold about as much as you can say in two seconds. That is why spans are longer for short words than long ones.

Getting it in

If attention lets something into short-term memory, what decides whether it lasts? Repeating it over and over, as you do with a phone number on the way to the phone, keeps it in short-term memory. Linking it to what you already know, so that it means something, helps it last.

In 1972 the psychologist Fergus Craik and a colleague put this as an idea about depth. Memory, they said, isn’t a matter of which box something reaches but of how deeply it was processed. You can notice only what a word looks like, or what it sounds like, or what it means, and the deeper you go, the better it lasts. In 1975 Craik and another colleague, Endel Tulving, put it to the test. People saw words and answered one question about each, about how it looked in print, how it sounded, or what it meant. Afterward, words whose meaning had been thought about were recognized far more often than words whose sound had been, and those far more than words judged only by their look.

The idea has a weakness. Nobody can measure depth on its own. If better memory is the sign of deeper processing, and deeper processing is the reason for better memory, the argument goes in a circle. Thinking about meaning also usually takes longer and more effort, so it is hard to say which of these does the work. And depth isn’t everything. In one experiment people who had attended to rhymes did better than those who had attended to meaning when the final test asked for rhymes. What you did when you learned something has to fit what you are asked when you need it.

Still, the plain lesson stands in today’s courses: thinking about what something means, and tying it to what you already know, helps it last. Memory champions use a version of this called the method of loci. They picture a walk through a familiar place and leave each thing to be remembered at a spot along the way, then walk the route again to collect them. Tested in a brain scanner, such champions turned out to have no unusual intelligence or brain structure. They were using a strategy, and ordinary people trained in it for six weeks improved sharply. In India, Vedic pandits train from childhood, for about ten years, to recite texts of 40,000 to 100,000 words exactly. Their brains show more gray matter than those of people matched to them, in areas used for memory and language, though a comparison like that can’t say whether the training made the difference.

Kinds of long-term memory

Long-term memory isn’t one thing either. Psychologists divide it first by whether you can put it into words.

Explicit memory, also called declarative memory, is what you can tell someone, and it comes in two kinds. Episodic memory is for events you lived through, with a when and a where: the morning of your SEE results, what you ate at a cousin’s wedding. Semantic memory is for facts and meanings, without the occasion you learned them on: that Kathmandu is the capital, that a triangle’s angles add up to 180 degrees, what a word means.

Implicit memory shows itself in what you do rather than in what you can say. Procedural memory is for skills, like riding a bicycle through a crowded lane or typing without looking, which are hard to explain to anyone and which come back even after years. Priming is when meeting something once makes it easier to take in the next time, even if you don’t remember meeting it: having seen a word recently, you read it a little faster. And the conditioned feelings of Chapter 8, a dog’s lifted ears at the rattle of a feeding bowl, are memories of this kind too. There is also memory for the future, called prospective memory: remembering to take your tiffin, or to ring your aunt at six.

The best evidence that these are really separate came from Henry Molaison, known as H.M., whose surgery is in Chapter 4. After it he could still hold six or seven digits for a moment, and a three-digit number for a quarter of an hour by repeating it to himself. So his short-term memory worked. What failed was storing anything new in long-term memory. His older memories were mostly kept, though tests found gaps in the years before the operation. And when he practiced drawing while seeing his hand only in a mirror, he got better over ten tries and three days, yet at the end had no memory of ever having done the task. His skills could still be learned. The memories he could tell about could not. One man’s brain had shown that short-term and long-term memory are different, and that within long-term memory, knowing that and knowing how are different too. One person proves little on his own, but many later patients, and many experiments, have pointed the same way.

Getting it out

Retrieval comes in two main kinds. Recall is pulling something out with little help, as when a question asks you to write what you know. Recognition is picking it out when you see it, as in a multiple-choice question. Recognition is usually easier, because the answer in front of you does part of the work. That is a clue to how retrieval works in general: through cues. A memory is found by following something linked to it, a smell, a tune, a place, a word.

So the place where you learned something ought to help you remember it, and in 1975 Alan Baddeley and a colleague tested this with a memorable experiment. Divers learned lists of words either on the shore or under about seven meters of cold sea off Scotland, then recalled them either in the same place or in the other one. Words learned underwater were remembered better underwater, and words learned on land better on land. The study is in almost every textbook. In 2021 a psychologist in Amsterdam repeated it with sixteen divers in a heated indoor pool, and found no advantage for recalling in the same place. He suggested reasons: the pool was shallow and warm, nothing like the open sea; all four tests were done in one day, not four; a television crew was filming. He concluded that more repeats are needed before the old study can be trusted as the example of the idea. So how much a whole surrounding place helps memory is less sure than the textbooks make it sound. The same is said of moods and bodily states, that a memory may come back more easily in the state it was made in.

The most useful discovery about retrieval is that doing it changes memory. In one experiment American students read short passages and then either read them again or tried to write down what they remembered, with no feedback. Five minutes later, rereading had done better. But two days or a week later, the students who had tested themselves remembered much more, even though the rereaders felt more confident they would remember. More than a hundred years of experiments, in laboratories and in classrooms, point the same way, and practice testing is rated among the most effective ways to study there are.

Spacing works on the same scale. A review in 2006 pulled together 839 comparisons in 317 experiments: studying something in sessions spread apart beats studying it for the same time all at once, and the longer you need to remember it, the wider the gaps that work best.

Why both work is still argued. Pulling a memory out seems to strengthen it directly, and a failed attempt also shows you what you don’t know yet. For spacing there are several explanations, none settled: a second look straight after the first feels familiar, so the mind gives it less attention; coming back later, in a different mood or room, ties the memory to more cues; and each return makes you pull up what you learned before, which is itself a kind of test. The effects are strong; the reasons are a live question. What to do about them, day to day, is in two short tips on this site, on testing yourself and spreading it out.

Nearly all of this was found with students in North America and Europe, and we found no memory study done in Nepal to set beside it.

So memory isn’t a single store but a set of processes. The senses hold nearly everything for a moment, attention passes a little on, and short-term or working memory holds about four chunks for half a minute while it works on them. What is linked to meaning goes into long-term memory, which has no known limit and comes in kinds, the facts and events you can tell and the skills you can only show. Getting it out works through cues, and each time you pull a memory out, you change it, which is why testing yourself and spacing your study work so well. SF’s 79 digits weren’t a bigger short-term memory, only a cleverer use of a long-term one, and that, in a small way, is open to anyone. The next chapter is about the other side: forgetting, and how a memory can feel certain and still be wrong.

Found a mistake? Please tell us.