Why your brain forgets Tuesday's vocab by Friday

In the 1880s, a German psychologist named Hermann Ebbinghaus did something a little unhinged. He invented thousands of meaningless syllables (WID, ZOF, that kind of thing), memorized them in long lists, and then tested how much he could recall at set intervals afterward. His only subject was himself. He kept it up for years.

It reads like a footnote from a strange century. But the curve he drew from that data became one of the foundations of memory science, and it's probably running your flashcard app right now.

A man and 2,000 nonsense syllables

Ebbinghaus used nonsense syllables on purpose. Real words come loaded with associations, and those associations make some words easy and others hard, which would muddy the measurement. Meaningless syllables gave him a clean baseline: raw memory, nothing to grab onto.

He'd learn a list to the point of perfect recall, then wait. Twenty minutes. An hour. A day. Six days. Each time, he measured how much effort it took to relearn the list, which told him how much had leaked out in between.

He called the measure savings. If a list took 10 minutes to learn on Monday and only 6 minutes to relearn on Wednesday, 40% of the original work had survived. It's a clever trick, because it picks up memory that's still in there but won't surface on its own, the way a song you'd swear you forgot comes back after one bar of the chorus.

Ebbinghaus's own savings data, 1885

Look at where the damage happens. He'd lost over 40% of his savings within an hour of learning. By day 2 the curve has mostly done its work, and the difference between 6 days and 31 days is small. Whatever survives the first stretch is comparatively safe.

Sit with that first number. The steepest forgetting of your Tuesday lesson happens before Tuesday's dinner, not sometime Thursday. Most learners aim their effort at the flat end of the curve, re-studying days later, when the cheap wins were in the first 24 hours all along.

The shape of forgetting

Here's the curve those numbers trace. The steepness in those first hours is the part that surprises people: a huge chunk of what you just learned is gone before you've had dinner. Then it levels out. What survives the first day tends to stick around a good while longer.

Retention after one study session

About that replication. In 2015, Murre and Dros re-ran the whole procedure: one dedicated 22-year-old subject, 70 lists of nonsense syllables learned and relearned over weeks of sessions, with modern controls and video-recorded protocol. The curve came back nearly identical to the 1885 original. 130 years apart, same shape, which is about as replicated as psychology gets.

They even reproduced the odd details, like the way the curve's early plunge outruns a simple exponential. Forgetting follows something closer to a power law: violent at first, then increasingly gentle, the same shape modern schedulers fit to review data from millions of learners. An 1885 self-experiment and a billion app reviews, agreeing.

Why the first hours are the worst

Two things gang up on a fresh memory.

First, it hasn't been filed yet. Consolidation, the slow process of moving a memory into stable long-term storage, takes hours to days, and a big share of it runs while you sleep. A word that doesn't make it through that filing run just drops. (This is why sleeping after you learn measurably helps, and why the late-night cram is doubly cursed.)

Second, interference. Everything you take in after the lesson (emails, a podcast, dinner conversation) lands on top of the fresh vocab and competes with it. Jenkins and Dallenbach showed this back in 1924: people forgot much less across a night of sleep than across the same number of waking hours, largely because sleep holds the incoming noise at bay.

There's one merciful asymmetry in the machinery: relearning is always cheaper than learning. That's what Ebbinghaus's savings measure was detecting. A word that's "gone" at the moment you fail it isn't gone at the roots, and one corrective review often restores in seconds what originally took minutes. Forgetting mostly hides memories rather than deleting them, and a good review catches them while they're still hiding shallow.

Interrupting the curve

Now the useful part. Every time you successfully pull a memory back before it's fully gone, you reset the curve, and the next decline is gentler. Review again, and it flattens more. Do this a handful of times, spaced right, and a word that would've evaporated by Friday will still be there next month.

Each review flattens the next drop

0 day

before an un-reviewed word starts slipping badly

0

well-timed reviews to move most words into long-term memory

0%

rough retention a scheduler can hold you at, if you show up

This is the entire logic behind spaced repetition, and it's why timing beats volume. Ten reviews crammed into tonight do less than four reviews landed at the right moments across three weeks. The when is doing most of the work.

And the right moments follow a rule of their own: the stronger the memory, the longer the ideal wait. Cepeda and colleagues (2008) mapped this across gaps from minutes to months, and the pattern is why review 1 comes tomorrow but review 5 comes in 2 months. Modern schedulers like FSRS fit that curve to every card you own.

The 50-year experiment

Here's the encouraging half of the story. Harry Bahrick tracked 733 people who'd studied Spanish in school, some of them 50 years earlier, and tested what remained (Bahrick, 1984). The losses came almost entirely in the first 3 to 6 years. Whatever survived that window then sat nearly unchanged for decades, even with no practice at all.

He called that stable state permastore. It's a real destination. Words you carry past the early cliff, with enough well-timed retrievals, have a genuine shot at staying for good, which reframes what your daily 20 minutes is actually buying: permanent residents, one word at a time.

Bahrick's data carries one more practical lesson. The people who reached permastore weren't the ones who'd studied most intensely; they were the ones whose learning had been spread across more years. Time-in-language beat effort-in-bursts, 5 decades out. It's the spacing effect again, drawn at the scale of a life.

Your Spanish isn't nonsense (mostly)

One honest footnote. Your deck is made of real words, and real vocabulary connects to things you already know, so it decays more slowly than Ebbinghaus's WIDs and ZOFs. A word like gato arrives with a cat attached.

But it follows the same shape, and the shape is what the scheduler models. So the plan writes itself. Learn the word. Let it fade just enough that pulling it back takes real effort. Pull it back. Repeat, with the gaps stretching each time. You don't have to track any of that by hand, which is the whole reason Verbamor exists, but now you know what it's doing under the hood.

---

Sources

Want more? Every claim on the site is traced to its paper on the research page.