What is the forgetting curve?

Ebbinghaus wanted to measure memory the way a physicist measures a quantity, which meant material without meaning or personal history. He built syllables out of a vowel between two consonants, strung them into series, and learned each series until he could recite it. He was the only subject. The work appeared in 1885 as Memory: A Contribution to Experimental Psychology.

To find out how much was left after a gap, he used the savings method. He learned the same series a second time and counted how much less work the second round took. Say the first learning took 1,000 seconds and the relearning 600: then 40% of the work was saved, and that 40% stood for what memory still held, even on a day when he could not recite a single syllable from the series.

The drop is steepest right after learning and slows down from there. Ebbinghaus put it in one line: forgetting is "very rapid at the beginning of the process and very slow at the end." Plotted on a chart, that gives the familiar shape, a sharp fall followed by a long, shallow tail.

Two readings of the numbers are easy to get wrong. They measure savings in relearning work, which answers a different question from how much you could recall on the spot. And they describe a series learned once and then left alone, with no review, no use and no link to anything Ebbinghaus cared about.

Jaap Murre and Joeri Dros repeated the experiment and published it in PLOS ONE in 2015. Dros, the second author, was the one subject. He spent 70 hours learning lists of nonsense syllables and relearning them after gaps from 20 minutes to 31 days. His curve came out very similar to the one from 1885, with one exception: his savings after 31 days were much lower than in the earlier curves.

Looking at Ebbinghaus's data and its replications together, the authors concluded that the curve is not completely smooth: it most probably jumps upwards at the 24-hour point. Research on sleep and memory would predict a jump like that, they write, though for this kind of experiment it is still to be established. They are also plain about the limit of studying one person: "it remains unclear what the shape of forgetting would be with other subjects."

The original curve belongs to material learned once and never touched again. Four things change its shape.

• Retrieval. Pulling an idea out of memory with the source out of sight slows the next decline more than reading it again. The testing effect guide goes through the studies.
• Relearning after a gap. Each round starts from what the last one saved, so it costs less. Katherine Rawson and John Dunlosky had 533 students practise recalling concepts, relearn them in later sessions, and sit a test one to four months afterwards. Relearning had pronounced effects on long-term retention for a fairly small cost in extra practice, and their advice was three correct recalls at first, then three relearning sessions at widely spaced intervals.
• Meaning. Ebbinghaus checked this himself with stanzas of Byron's Don Juan, about 80 syllables each. A stanza took him hardly nine repetitions, where he reckoned a series of 80 to 90 nonsense syllables would have needed 70 to 80.
• Sleep, possibly. The upward jump Murre and Dros found at 24 hours fits what sleep research predicts, though the authors stop short of saying sleep caused it.

• One subject, who was also the experimenter and knew what he was testing for.
• Syllables chosen for having no meaning, so the curve leaves out the thing that helps memory most.
• A savings measure, sensitive enough to find traces of a series he could no longer recite. It tells you how much easier relearning is, which differs from whether you can say the answer now.
• Material learned once and then ignored. Most of what you learn in daily life is met again, used and linked to other ideas.

The curve is a sound description of how fast unrehearsed, meaningless material fades for one careful learner, and the 2015 replication says the description holds. Charts that put a precise figure on how much everyone forgets by tomorrow ask more of it than it can give.

scrollandlearn turns the posts you save from 15+ platforms into short lessons, filed into topics. A saved post is the textbook case of material met once and left alone, so the app adds the step the curve asks for: Today's quiz brings the lesson back, you answer its question from memory, then grade yourself Forgot, Kind of or Knew it.

Forgot brings a lesson back tomorrow and starts it over, Kind of brings it back in 4 days, and Knew it in 14 days, then 35, 88 and 219 days, then a year. Forgot treats the lesson as back at the steep start of the curve, so it returns the next day. Kind of keeps the progress the lesson had and checks again in 4 days. Each Knew it in a row stretches the gap, in line with the relearning research above, where rounds spread widely apart paid off for months.

The gaps come from your grade and from how many times in a row you have known the lesson. The app does not estimate a personal forgetting curve for you, so an honest grade is what keeps the timing right.

The example illustrates the lesson a LinkedIn post about CV writing becomes. Its question asks how each line of a CV should start. Grade it Knew it and it returns in 14 days; know it again and the gap grows to 35 days, then 88. Miss it once and it comes back tomorrow and climbs from the start.