How We Learn

by Benedict Carey

  There is an emerging theory that accounts for these and related ideas. It’s called the New Theory of Disuse, to distinguish it from an older, outdated principle stating, simply, that memories evaporate entirely from the brain over time if they’re not used. The new theory is far more than an updating, though. It’s an overhaul, recasting forgetting as the best friend of learning, rather than its rival.

  A better name for it, then, might be the Forget to Learn theory. That phrase captures its literal implications and its general spirit, its reassuring voice. One implication, for instance, is that forgetting a huge chunk of what we’ve just learned, especially when it’s a brand-new topic, is not necessarily evidence of laziness, attention deficits, or a faulty character. On the contrary, it is a sign that the brain is working as it should.

  No one knows why we should be such poor judges of forgetting or other mental skills that are so indispensable, so automatic, that they feel deeply familiar. Yet we are. And it helps to count the ways.

  • • •

  Let’s go back to the beginning, then. Let’s go back to the first learning laboratory of them all, to its sole occupant, and his most important contribution—the Forgetting Curve. The Forgetting Curve is exactly what it sounds like, a graph of memory loss over time. In particular, it charts the rate at which newly learned information fades from memory. It’s a learning curve, turned upside-down:

  This curve, first published in the late 1880s, falls well short of breathtaking. It’s what anyone might draw if asked to guess how memory changes with time. Yet its creator, Hermann Ebbinghaus, wasn’t one for idle guesswork. He was exacting by nature, compulsive about evidence. He had to be, given his ambitions. In the late 1870s, as a young philosophy Ph.D., he zigzagged through Europe, thinking big. He longed to bridge philosophy and science, to apply rigorous measurement to some aspect of human nature or psychology. The only problem was, he didn’t know where to start. He was poking around in a secondhand Paris bookstall one afternoon when he pulled from the shelf a volume called Elements of Psychophysics by Gustav Fechner. A scientist with a mystical bent, Fechner saw a unifying mathematical connection between the inner, mental world and the outer, natural one. He argued that every human experience, even one as ephemeral as memory, should be reducible to measurable units that could be plugged into an equation of some sort. Fechner’s reputation as a scientist—he’d done elegant experiments on the sensation of touch—lent his more grandiose ideas some weight.

  As he read, Ebbinghaus felt something inside him shift—a sensation he would describe, years later, to a student. He must have glimpsed his future as well, right then and there, because he later dedicated his greatest work, Memory: A Contribution to Experimental Psychology, to Fechner.

  The memory equation. Did it even exist? If so, could it be written down?

  Memories come in so many shapes and sizes. There are the hour-long and the lifelong; there are dates and numbers, recipes and recitals; not to mention stories, emotional perceptions, the look on a child’s face when he’s dropped at the bus stop on the first day of school, the knowing smile shared between two friends who think no one is looking: the tapestry of hijinks and heartbreaks that make up a life. Our ability to recall specific facts also varies widely. Some people are good with names and faces; others are much better at retrieving numbers, dates, formulas. How on earth do you measure such a shape-shifting ghost, much less study it?

  A generation of scientists before Ebbinghaus had essentially stood down, taking a pass on the question. It was too much. The variables were overwhelming.

  Yet where some saw a justified caution, Ebbinghaus saw a lack of nerve. “At the very worst we should prefer to see resignation arise from the failure of earnest investigations rather than from the persistent, helpless astonishment in the face of the difficulties,” he wrote, in explaining his motives for pursuing the memory equation. He would take the dare if there was no one else. He reasoned from first principles. To study how the brain stores new information, he needed information that was, in fact, new. A list of nouns or names or numbers simply wouldn’t do; people walk around with an enormous storehouse of associations for all of these things. Even abstract sketches have a Rorschach-like, evocative quality. Stare long enough at a cloud and it begins to look like a dog’s head, which in turn activates hundreds of dog-related circuits in the brain. Our brain can impute meaning to almost anything.

  How Ebbinghaus arrived at his solution remains a mystery. “Was it an invention in the commonly accepted sense of the term, that is to say, deliberate?” wrote the American psychologist David Shakow, much later, in a biographical essay. “Or was it largely a discovery? What part did the gurgle of an infant, a transient progression to infancy, the reading of Jabberwocky, the expletives of the Paris coachman for the London cabbie, play?”

  What Ebbinghaus created was a catalog of nonsense sounds. These were single syllables, formed by sticking a vowel between two consonants. RUR, HAL, MEK, BES, SOK, DUS. By and large, they were meaningless.

  Ebbinghaus had found his generic memory “units.”

  He created about 2,300 of them—a pool of all possible syllables, or at least as many as he could think of. He put together lists of the syllables, random groupings of between seven and thirty-six each. Then he began to memorize one list at a time, reading the syllables out loud, pacing himself with a metronome, keeping track of how many repetitions he needed to produce a perfect score.

  By the time he landed a job as an instructor, at the University of Berlin in 1880, he’d logged more than eight hundred hours of practice with his nonsense sounds. He continued the work in his small office, pacing the floor, a compact, bushy-bearded man in Ben Franklin spectacles, spitting out the syllables at a rate of as many as 150 a minute. (In another era or another country, he might have been hauled off and fitted with a lunatic suit.) He tested himself at various intervals: Twenty minutes after studying. An hour. A day later, then a week. He varied the duration of his practice sessions, too, and found (surprise) that more practice sessions generally resulted in higher test scores and a slower rate of forgetting.

  In 1885, he published his results in Memory: A Contribution to Experimental Psychology, describing a simple way to calculate the rate of forgetting after a study session. The equation wasn’t much to look at, but it was the first rigorous principle in the emerging branch of psychology—and precisely what he’d set out to find a decade earlier in that Paris bookstall.

  Ebbinghaus had his equation (others would plot it as a graph).

  He hadn’t changed the world. He did, however, launch the science of learning. “It is not too much to say that the recourse to nonsense syllables, as a means to the study of association, marks the most considerable advance in this chapter of psychology since the time of Aristotle,” wrote English scientist Edward Titchener a generation later.

  Ebbinghaus’s Forgetting Curve captured the minds of many theorists and would not let go. In 1914, the influential American education researcher Edward Thorndike turned Ebbinghaus’s curve into a “law” of learning. He called it the Law of Disuse, which asserted that learned information, without continued use, decays from memory entirely—i.e., use it or lose it.

  The law felt right. It certainly seemed to square with experience, defining how most people thought of learning and to this day still do. Yet that definition hides more than it reveals.

  • • •

  Here’s an at-home exercise that is simple, painless, and full of literary nutrition. Take five minutes and study the verse below. Read it carefully and try to commit it to memory. It’s from the poet Henry Wadsworth Longfellow’s “The Wreck of the Hesperus.”

  At daybreak, on the bleak sea-beach,

  A fisherman stood aghast,

  To see the form of a maiden fair,

  Lashed close to a drifting mast.

  The salt sea was frozen on her breast,

  The salt tears in her eyes;

  And he saw her hair, like the brown s
ea-weed,

  On the billows fall and rise.

  Such was the wreck of the Hesperus,

  In the midnight and the snow!

  Christ save us all from a death like this,

  On the reef of Norman’s Woe!

  Okay, now put the book aside and make a cup of coffee, take a walk, listen to the news. Distract yourself for about five minutes, the same amount of time you took to study it. Then, sit and write down as much of the poem as you can. Save the result (you’ll need it later).

  This is exactly the test that an English teacher and researcher named Philip Boswood Ballard began administering to schoolchildren in the early 1900s in London’s working-class East End. The children were thought to be poor learners, and Ballard was curious to find out why. Was it a deficit of initial learning? Or did something happen later that interfered with recall? To find out, he had them study various material, including ballads like Longfellow’s, to see if he could pinpoint the source of their learning problems.

  Only the children had no obvious learning deficits that Ballard could find. On the contrary.

  Their scores five minutes after studying were nothing special. Some did well and others didn’t. Ballard wasn’t finished, however. He wanted to know what happened to the studied verse over time. Did memory somehow falter in the days after these children studied? To find out, he gave them another test, two days later. The students were not expecting to be retested and yet their scores improved by an average of 10 percent. Ballard tested them once more, again unannounced, days later.

  “J.T. improved from 15 to 21 lines in three days,” he wrote of one student. “Imagined she saw the lines in front of her.” Of another, who improved from three to eleven lines in seven days, he remarked: “Pictured the words on the blackboard (the poetry in this case was learnt from the blackboard).” A third, who recalled nine lines on the first test and, days later, thirteen, told Ballard, “as I began to write it, I could picture it on the paper before me.”

  This improvement wasn’t merely odd. It was a flat contradiction of Ebbinghaus.

  Ballard doubted what he was seeing and ran hundreds of additional tests, with more than ten thousand subjects, over the next several years. The results were the same: Memory improved in the first few days without any further study, and only began to taper off after day four or so, on average.

  Ballard reported his findings in 1913, in a paper that seems to have caused mostly confusion. Few scientists appreciated what he’d done, and even today he is little more than a footnote in psychology, a far more obscure figure than Ebbinghaus. Still, Ballard knew what he had. “We not only tend to forget what we have once remembered,” he wrote, “but we also tend to remember what we have once forgotten.”

  Memory does not have just one tendency over time, toward decay. It has two.

  The other—“reminiscence,” Ballard called it—is a kind of growth, a bubbling up of facts or words that we don’t recall having learned in the first place. Both tendencies occur in the days after we’ve tried to memorize a poem or a list of words.

  What could possibly be going on?

  One clue comes from Ebbinghaus. He had tested memory using only nonsense syllables. The brain has no place to “put” these letter trios. They’re not related to one another or to anything else; they’re not part of a structured language or pattern. The brain doesn’t hold on to nonsense syllables for long, then, because they are nonsense. Ebbinghaus acknowledged as much himself, writing that his famous curve might not apply to anything more than what he had studied directly.

  Forgetting, remember, is not only a passive process of decay but also an active one, of filtering. It works to block distracting information, to clear away useless clutter. Nonsense syllables are clutter; Longfellow’s “The Wreck of the Hesperus” is not. The poem may or may not become useful in our daily life, but at least it is nested in a mesh of neural networks representing words and patterns we recognize. That could account for why there would be a difference in how well we remember nonsense syllables versus a poem, a short story, or other material that makes sense. Yet it does not explain the increase in clarity after two days without rehearsal, the “salt tears” and “hair like brown sea-weed” floating up from the neural deep. Those “slow” East Enders showed Ballard that remembering and forgetting are not related to each other in the way everyone assumed.

  The Forgetting Curve was misleading and, at best, incomplete. It might even need to be replaced altogether.

  • • •

  In the decades after Ballard published his findings, there was a modest flare of interest in “spontaneous improvement.” The effect should be easy to find, scientists reasoned, in all kinds of learning. Yet it wasn’t. Researchers ran scores of experiments, and the results were all over the place. In one huge 1924 trial, for instance, people studied a word list, and took a test immediately afterward. They were then given a follow-up test, after varying delays: eight minutes, sixteen minutes, three days, a week. They did worse over time, on average, not better.

  In a 1937 experiment, subjects who studied nonsense syllables showed some spontaneous improvement after an initial exam—but only for about five minutes, after which their scores plunged. A widely cited 1940 study found that people’s recall of a set of words, a set of brief sentences, and a paragraph of prose all declined over a twenty-four-hour period. Even when researchers found improvement for one kind of material, like poetry, they’d find the opposite result for something else, like vocabulary lists. “Experimental psychologists began to tinker with Ballard’s approach and, as if struggling in quicksand, became progressively mired in confusion and doubt,” wrote Matthew Hugh Erdelyi, of Brooklyn College, in his history of the era, The Recovery of Unconscious Memories.

  The mixed findings inevitably led to questions about Ballard’s methods. Were the children he tested really recalling more over time, or was their improvement due to some flaw in the experimental design? It wasn’t a rhetorical question. What if, for example, the children had rehearsed the poem on their own time, between tests? In that case, Ballard had nothing.

  In an influential review of all published research up through 1943, one British learning theorist, C. E. Buxton, concluded that Ballard’s spontaneous improvement effect was a “now-you-see-it-now-you-don’t phenomenon”—in other words, a phantom. It wasn’t long before many scientists followed Buxton’s lead and begged off the hunt. There were far better things to do with the tools of psychology than chase phantoms, and certainly more culturally fashionable ones.

  Freudian therapy was on the rise, and its ideas of recovered memories easily trumped Ballard’s scraps of Longfellow for sex appeal. The two men’s conceptions of recovery were virtually identical, except that Freud was talking about repressed emotional trauma. Excavating those memories and “working through” them could relieve chronic, disabling anxiety, he claimed. It could change lives. If those were phantoms, they were far more lifelike than a heap of recited poetry.

  Besides, the real juice in learning science by the middle of the century was in reinforcement. It was the high summer of behaviorism. The American psychologist B. F. Skinner showed how rewards and punishments could alter behavior, and accelerate learning in many circumstances. Skinner tested various reward schedules against one another and got striking results: An automatic reward for a correct answer leads to little learning; occasional, periodic rewards are much more effective. Skinner’s work, which was enormously influential among educators, focused on improving teaching, rather than on the peculiarities of memory.

  Yet Ballard’s findings didn’t disappear completely. They continued to marinate in the minds of a small group of psychologists who couldn’t shake the idea that something consequential might be slipping through the cracks. In the 1960s and 1970s, these curious few began to separate the poetry from the nonsense.

  The Ballard effect was, and is, real. It was not due to an experimental design flaw; the children in his studies could not have rehearsed lines
that they did not remember after the first test. You can’t practice what you don’t remember. The reason researchers had had so much trouble isolating Ballard’s “reminiscence” was because the strength of this effect is highly dependent on the material being used. For nonsense syllables, and for most lists of vocabulary words or random sentences, it’s zero: There’s no spontaneous improvement on test scores after a day or two. By contrast, reminiscence is strong for imagery, for photographs, drawings, paintings—and poetry, with its word-pictures. And it takes time to happen. Ballard had identified the “bubbling up” of new verse in the first few days after study, when it’s strongest. Other researchers had looked for it too early, minutes afterward, or too late, after a week or more.

  Matthew Erdelyi was one of those who was instrumental in clarifying reminiscence, and he began by testing a junior colleague, Jeff Kleinbard, then at Stanford University. Erdelyi gave Kleinbard a group of forty pictures to study in a single sitting, on the pretext that he “should have the experience of being a subject” before conducting experiments of his own. In fact, he was a subject, and Erdelyi tested him repeatedly, without warning, over the following week. The results were so clear and reliable—Kleinbard remembered increasingly more on tests over the first two days—that the two of them set up larger studies. In one, they had a group of young adults try to memorize a series of sixty sketches. The participants saw the sketches one at a time, projected on a screen, five seconds apart: simple drawings of things like a boot, a chair, a television.

  The group took a test right after and tried to recall all sixty, in seven minutes, writing down a word to describe each sketch recalled (the sketches had no accompanying words). The average score was 27. Ten hours later, however, their average was 32; a day later, 34; by four days, it was up to 38, where it plateaued. A comparison group, who studied sixty words presented on slides, improved from 27 to 30 in the first ten hours—and no more. Their scores slipped slightly over the next several days. Soon it was beyond dispute that memory, as Erdelyi put it in a recent paper, “is a heterogeneous, mottled system that both improves and declines over time.”