How memory works

Before anything is remembered, the senses hold a brief copy of what they have just taken in. In 1960 George Sperling flashed a grid of twelve letters at people for a twentieth of a second. They could name only four or five. Had they seen so few? He changed one thing. Just after the flash, a tone told them which of the three rows to report. Now they got about three of the four letters in whichever row was asked for. Most of the grid had been there for a moment, and it faded while they were still naming the first few letters.

This fleeting store is sensory memory. The visual kind, called iconic memory, fades within about half a second. The hearing kind, echoic memory, lasts up to about four seconds, which is how you can replay the end of a sentence you were not listening to.

What you attend to passes from the senses into short-term memory, which is small and leaky. In 1956 George Miller put its span at about seven items, give or take two. Later work, which stopped people grouping the items, puts it nearer four. Unless you keep repeating them to yourself, which is called rehearsal, most are gone within twenty seconds.

There is a way round the limit. Twelve digits are too many to hold, but grouped as three familiar years they are three items. Packing small items into larger meaningful units is called chunking, and it is why phone numbers are written in blocks.

Psychologists now prefer the name working memory, which took hold after a 1974 paper by Alan Baddeley and Graham Hitch. The name says that information is used here and not just kept: a sum, a sentence or a route, held while you act on it.

Long-term memory, by contrast, has no known limit, and what it holds can last a lifetime. It is not one thing, though. Start with the memories you can bring to mind on purpose, aware that you are remembering, and put into words. These are called declarative, or explicit, memories, and they come in two kinds.

One is your record of events you lived through: your first day at school, what you ate this morning. Each is tied to a time and a place, and you can travel back to it in your mind. This is episodic memory. The other is your store of general knowledge: that Cairo is in Egypt, what a triangle is, what a word means. You know these things without recalling when or where you learned them. This is semantic memory.

In 1953 an American surgeon tried to cure a young man’s severe epilepsy by removing tissue from both sides of his brain, including most of a structure called the hippocampus. The seizures eased. But the patient, Henry Molaison, long known only as H.M., could no longer form lasting memories of new facts or events. He would greet as a stranger a doctor he had met many times. Losing the power to lay down new memories is called anterograde amnesia.

One thing surprised the psychologist Brenda Milner. She asked him to trace a star while watching his hand in a mirror, a hard skill. Over three days he grew better, yet each day he said he had never done it before. Memory for how to do things, called procedural memory, was intact. It is one kind of implicit memory: memory that shows in what you do, with no awareness of remembering.

What decides whether something met once will stick? In 1975 Fergus Craik and Endel Tulving showed people words one at a time, with a quick question about each. Some questions were about its look: is it in capital letters? Some were about its sound: does it rhyme with train? Some were about its meaning: would it fit in this sentence? Nobody expected a test. Later, among the words whose answer had been yes, people recognised 15 per cent of those judged by look and 81 per cent of those judged by meaning.

Thinking about meaning ties a new item to what you already know, and each tie is a route back to it. Building such ties on purpose is called elaboration. Orators of ancient Greece and Rome learned speeches by tying each point to a place on an imagined walk through a building they knew, now called a memory palace.

A stored memory is found by way of something that points to it, a retrieval cue. A smell can bring back a whole childhood kitchen. A first letter can release a name. The cues that work best are the ones that were present when the memory was laid down, because they were stored along with it.

That includes the place itself. Memory that is better where it was formed is called context-dependent memory. In the best-known test, in 1975, divers learned word lists underwater or on the beach and recalled more when tested in the same setting. A careful repeat in 2021 did not find the difference. Across dozens of other experiments, though, a modest advantage of matching surroundings does hold up. The practical lesson is to build cues that travel with you, such as a question that leads to its answer, and not to rely on the room.

How fast do memories fade? The first person to measure it was Hermann Ebbinghaus, a German psychologist who in the 1880s used himself as the subject. He learned lists of meaningless syllables, then learned each list again after twenty minutes, an hour, a day, a month. The less work the second learning took, the more of the list he had kept.

His results trace a line called the forgetting curve. It falls steeply at first: by this measure, more than half of what he had learned was gone within an hour. Then it flattens, so that some was still left a month later. The lesson is uncomfortable. Most forgetting happens soon after learning, long before you would think of revising. But the curve describes material met once and then left alone. Each time something is recalled and relearned, the fall that follows is slower.

The passing of time is not the whole story of forgetting. Often the trouble is other memories. When you learned something similar before or after, the two compete, and the wrong one may come to mind, or neither. Forgetting caused by competing memories is called interference.

It runs in two directions. Move house, and for weeks your hand writes the old postcode: earlier learning is getting in the way of the new. This is proactive interference, the old reaching forward. Months later you are asked for the old postcode and can think only of the new one: later learning has blocked the earlier. This is retroactive interference, the new reaching back. Both are strongest between things that resemble each other, which is why two similar languages studied together are easily muddled.

Read someone a list of twenty words and ask for them back in any order. Their chance of recalling a word depends on where it stood. In a 1962 experiment, people recalled the last word 95 times in 100, the first 46 and the tenth only 22. The advantage of words near the start is the primacy effect, and that of words at the end is the recency effect. The middle of the list fares worst.

The two ends are remembered for different reasons, and a 1966 experiment pulled them apart. When people had to count aloud for thirty seconds before recalling, the recency effect vanished and the primacy effect stayed. The last words had only been sitting in short-term memory, and the counting pushed them out. The first words had been rehearsed most, and had already reached long-term memory.

It is natural to picture memory as a recording that plays back. The British psychologist Frederic Bartlett showed in 1932 that it is nothing of the kind. He had English students read a Native American folk tale, The War of the Ghosts, full of events and customs strange to them, and retell it weeks or months later. The story changed in a regular way. It grew shorter and tidier. Canoes became boats. A hunt for seals became a fishing trip. Unfamiliar parts were reshaped into familiar ones.

Bartlett’s explanation was that we do not store events whole. We keep the gist and rebuild the rest each time from a schema, a mental framework of what such things are usually like. Memory, he concluded, is reconstructive. Each recall is a fresh rebuilding, and it can alter what is recalled next time.

Can a rebuilt memory be steered? In 1974 Elizabeth Loftus and John Palmer showed people films of car accidents and asked how fast the cars were going. For some the question said the cars hit each other, for others that they smashed into each other. That one word moved the average estimate from 34 to 41 miles an hour. In a second experiment, a week later, about twice as many of the smashed group remembered broken glass, though the film showed none. Changing a memory through information supplied after the event is called the misinformation effect.

Vividness is no protection. The day after the space shuttle Challenger exploded in 1986, students wrote down how they heard the news. Two and a half years later many gave a different account, and were confident of it. Sharp, sure memories of shocking news are called flashbulb memories, and they drift like any other.

How can anyone rely on memory when a law or a life depends on it? Not by asking how sure the witness feels. The scholars who gathered hadith, the reports of what the Prophet said and did, faced this problem from the eighth century on. Their answer was to test the person remembering.

A narrator had to be upright, and also precise: known to pass on accurately what he heard. They checked precision by comparing his reports with those of others known to be reliable, above all fellow students of the same teacher. Yaḥyā ibn Maʿīn, a ninth-century critic of Baghdad, heard one teacher’s books from seventeen of his students, to tell the teacher’s mistakes from theirs. A narrator whose reports often differed from the others’ was not relied on, however pious he was. When a narrator’s memory failed in old age, only his earlier teaching was accepted.

Everything above points to a few habits that make memories last. None of them is what most people do by instinct.

The first is retrieval practice: testing yourself instead of rereading. Pulling an answer out of memory strengthens the route to it far more than looking at it again, which is why this reading stops to ask you questions. The second is the spacing effect. The same study time works better spread over days than packed into one sitting, because each return catches the memory as it begins to fade. The third costs no effort at all. During sleep the brain goes back over the day’s new memories and helps to make them stable, a process called consolidation. A lesson followed by a night’s sleep is kept better than one followed by the same hours awake.