Mark ở vũ trụ khác · Another Universe Papers That Changed How We Learn #1 · Memory · Review

The Forgetting Curve

reviewreview

Hermann Ebbinghaus · Über das Gedächtnis (On Memory) · Leipzig, 1885

The papera monograph on memory, printed in 1885, translated into English in 1913
Findingwe forget fastest at first, then more slowly; spreading reviews over days beats cramming
ReplicatedHeller, Mack and Seitz (1991); Murre and Dros (PLOS ONE, 2015)
LabelSolid, as long as the numbers are read correctly

Source: Ebbinghaus (1885), English translation by Ruger and Bussenius (1913); Heller, Mack and Seitz (1991), Zeitschrift für Psychologie 199; Murre and Dros (2015), PLOS ONE 10(7): e0120644. The label is Mark's assessment.

Re-read in 1 minute

Essence

One person, measuring himselfEbbinghaus was both the scientist and the only participant.
2,300 nonsense syllablesConsonant, vowel, consonant, with no meaning, so memory could be measured "clean".
Fast, then slowAfter 20 minutes he still saved 58% of relearning effort. After 1 day, 34%. After 31 days, still 21%.
Spread it out38 readings over 3 days matched 68 readings crammed into 1 day.
Not "we forget 70%"His numbers measure relearning effort saved, not the percentage still remembered.

Source: Ebbinghaus (1885/1913), chapters 3, 7, 8; syllable count per Britannica and the APS Observer (secondary source). How to read "savings": Murre and Chessa (2023), Psychonomic Bulletin & Review.

Part 1

The paper's question

Can memory be measured the way temperature is? In 1885 a German answered yes: we forget along a curve with a clear shape, and how we review decides how steep that curve is.

The question is Mark's summary of the aim Ebbinghaus states in the preface and chapter 1.

Step 1Make "clean" materialNonsense syllables, so nothing already in the head helps.
→
Step 2Measure by relearning timeLearn, wait, relearn, and see how much time is saved.
→
Step 3Draw the curve7 points from 20 minutes to 31 days. Plus a test of massed versus spread practice.
Part 2

The story of the discovery

Who, where, how, and why it was bold at the time.

1850 · Barmen, GermanyHermann Ebbinghaus is born. He studies philosophy; he is not yet a psychologist in today's sense.
1870s · A second-hand bookHe reads Gustav Fechner's Elemente der Psychophysik, a book that measures sensation with numbers. The thought: if sensation can be measured, why not memory? Sources disagree on whether he bought it in Paris or London.
1879–1880 · About 7 months of self-testingHe memorises lists of nonsense syllables to the beat of a metronome, then relearns them after different delays, timing every session.
1884 · Trying poetryA comparison with meaningful verse: six stanzas of Byron's Don Juan.
1885 · Über das GedächtnisPrinted in Leipzig. One of the first times human memory was studied with measured experiments.
1913 · English editionRuger and Bussenius translate it as Memory: A Contribution to Experimental Psychology.

Source: birth year and the Fechner book per Wikipedia, New World Encyclopedia, Encyclopædia Universalis (secondary sources, which conflict on where he bought the book); 1879–1880 per the documentation of the Ebbinghaus dataset in the R package HistData (Friendly, data supplied by Jaap Murre); Don Juan per the 1913 translation, chapter 5.

How "savings" is measured. The first learning takes a long time. After a delay, relearning the same list takes less. The time saved, as a percentage of the first session, is the savings score. Illustrative example: first 100 seconds, later 66 seconds, savings 34 percent. how Ebbinghaus measured memory: "savings" first learning 100 seconds to the first error-free recital relearning a day later 66 seconds 34 seconds saved savings = (100 − 66) / 100 = 34% Illustrative round numbers. "34%" is relearning effort saved, not the percentage remembered.
Measuring even what we think is gone. Often we cannot recall a thing, yet relearning it is still faster. This measure captures that "hidden" memory.Source: definition in Murre and Chessa (2023), Psychonomic Bulletin & Review; the "repetitions up to the first errorless reproduction" criterion per the 1913 translation, chapter 3. The numbers in the figure are Mark's illustration.

Bold for its time

Before him, memory belonged to philosophy and was argued over. Ebbinghaus put it into a measured experiment.

Remarkable discipline

Reading at a steady beat and timing every learning and relearning session, so the measurements could be compared.

A weakness he admitted

Only one person. The introduction to the 1913 translation says the numerical results are therefore "limited in significance".

Source: introduction to the 1913 translation (via excerpts in search results); metronome per the APS Observer and Wikipedia (secondary sources).

Part 1 of 3

The discovery

one person2,300 syllablessavingsquiz
Part 2 of 3

What it says, how it was received

7 points68 vs 38the 70% myth
Part 3 of 3

New research, lessons, Mark's score

2015 replication317 experimentsreview plan
Part 3

What the paper says

In 3 sentences, then in pictures.

  1. Memory can be measured: how much relearning time drops tells you how much was kept.
  2. We forget fastest right after learning, then more slowly, but some of it is still there after a month.
  3. For the same number of readings, spreading them over several days keeps more than massing them at once.

The three sentences are Mark's summary of chapters 7 and 8.

Ebbinghaus's forgetting curve, measured as percent savings on relearning: 20 minutes 58.2; 1 hour 44.2; about 9 hours 35.8; 1 day 33.7; 2 days 27.8; 6 days 25.4; 31 days 21.1. The horizontal axis is not evenly spaced. 60%30%0 58.244.235.833.727.825.421.1 20 min1 hour~9 hours1 day2 days6 days31 days % of relearning effort saved (higher means more retained) uneven horizontal axis: the points are minutes to weeks apart
A steep drop in the first hour, then almost flat. From 20 minutes to 1 hour, 14 points go. From 6 days to 31 days, just over 4.Source: Ebbinghaus (1885/1913), chapter 7, results table; the values 0.582 (20 min), 0.442 (1 hour), 0.211 (31 days) match the comparison table in Murre and Dros (2015); "about 9 hours" is given as 8.8 hours in some sources.
Massed versus spread practice. For a 12-syllable list, 68 readings crammed into the day before gave the same relearning result as 38 readings spread over the three days before. same result when relearning a 12-syllable list massed in 1 day 68 spread over 3 days 38 the three purple blocks are three days; the even split is illustrative, the book gives no per-day count
Almost half the readings, same result. This is the earliest record of what we now call the spacing effect.Source: Ebbinghaus (1885/1913), chapter 8 (psychclassics.yorku.ca, memory8.htm, checked via excerpts); also quoted in the APS Observer and PMC3390592.
With any considerable number of repetitions a suitable distribution of them over a space of time is decidedly more advantageous than the massing of them at a single time.Ebbinghaus (1885), 1913 translation, chapter 8, p. 89; quoted via db.barbanon.org (secondary source).
More: how much easier is poetry than nonsense?

In 1884 Ebbinghaus tried six stanzas of Don Juan. All six took 52 readings, hardly nine per stanza. With nonsense syllables, according to one summary, a 7-syllable list needed a single reading, but 16 syllables needed 30 and 24 syllables needed 44.

He noted himself that this comparison was not made under the same conditions, so no exact ratio should be drawn. The point worth keeping: meaning helps memory enormously, which is exactly why he removed it when measuring.

Source: 1913 translation, chapter 5 (Don Juan, checked via excerpts); readings by list length per SuperMemo (secondary source). The online claim that poetry took "one tenth of the effort" was not found in the original and is not used.

Part 4

How this paper has been judged

Where it appeared, citations, replications, criticism. Each line has a source.

Where, whenMonograph, Duncker & Humblot, Leipzig, 1885. English edition: Teachers College, Columbia University, 1913.
CitationsNot retrieved. Google Scholar, Semantic Scholar and OpenAlex were all blocked from the writing environment (checked 15/10/2026). No number is given.
Replicated?Yes. Heller, Mack and Seitz (1991), two people, in Germany. Murre and Dros (2015), one person, about 70 hours of learning over 75 days: "the Ebbinghaus forgetting curve has indeed been replicated".
Large synthesisRubin and Wenzel (1996) collected 210 published data sets: the "fast then slow" shape recurs everywhere, but several mathematical functions fit almost equally well.
Main criticism(1) One person, testing himself. (2) Bartlett (1932, Remembering): stripping out meaning is a remedy "at least as bad as the disease", and says little about everyday memory. (3) The numbers hold only for nonsense syllables and the savings measure.
StandingIn 1985, the centenary, Roediger (Contemporary Psychology) called him one of psychology's foremost pioneers; the Journal of Experimental Psychology: Learning, Memory, and Cognition ran Murdock's piece on his contributions. Awards: no source found, none listed.
VerdictSOLID The shape of the curve and the benefit of spreading practice have been replicated. Read it correctly: the number is "relearning effort saved", measured on nonsense syllables.

Source: HistData documentation (friendly/HistData, man/Ebbinghaus.Rd) for Heller et al. (1991) and Dros's 75 days; abstract of Murre and Dros (2015), PLOS ONE; Rubin and Wenzel (1996), Psychological Review 103(4), 734–760; Bartlett quoted via J. F. Kihlstrom's Berkeley notes (secondary source); Roediger (1985), Contemporary Psychology 30(7), 519–523; Murdock (1985), JEP: LMC 11(3), 469–471. The label is Mark's assessment.

Part 5

Fact-check: what holds up and what doesn't

14 things often said about Ebbinghaus and the forgetting curve. Tap a row for the reason.

SOLIDEbbinghaus was his own and only participant

He ran the tests and was the one tested. Encyclopedia.com and the HistData documentation both say so (secondary sources).

SOLIDHe used about 2,300 nonsense syllables

Britannica, Wikipedia and the APS Observer all give about 2,300 consonant-vowel-consonant syllables (secondary sources, consistent with each other).

SOLIDWe forget fastest just after learning, then more slowly

Table in chapter 7; replicated by Heller et al. (1991) and Murre and Dros (2015); Rubin and Wenzel (1996) found this shape in 210 data sets.

SOLIDAfter 1 day savings were 33.7%, after 31 days still 21.1%

Chapter 7 table. The 31-day value (0.211) matches the comparison table in Murre and Dros (2015). Note: one person's numbers, for nonsense syllables.

SOLIDThe curve was replicated in 2015

Murre and Dros (PLOS ONE, 2015): one person learned for about 70 hours, relearned after 20 minutes to 31 days, and concluded the curve "has indeed been replicated".

SOLIDSpreading reviews over several days beats cramming

Chapter 8: 38 readings over 3 days matched 68 in 1 day. More than 100 years later, Cepeda et al. (2006) synthesised 317 experiments and confirmed the benefit of spacing.

SOLIDA single participant is the original study's big weakness

The introduction to the 1913 translation itself says the numerical results are "limited in significance". That is why replications with other people matter.

CAUTIONThe jump at the 24-hour point is caused by sleep

Murre and Dros write that the curve "most probably shows a jump upwards starting at the 24 hour data point". Sleep is a plausible explanation, but a one-person study cannot isolate it.

CAUTIONThe forgetting curve is an exact exponential

Rubin and Wenzel (1996) tested 105 functions: logarithmic, power and a few other forms fit almost equally well and are hard to tell apart. Wickens (1998) commented further. "Exponential" is only an approximation.

CAUTIONThe "1, 3, 7, 30 days, after Ebbinghaus" review schedule is his

Chapter 8 only compares massed and spread practice; it gives no schedule. Spaced-repetition systems arrived about 100 years later (supermemo.guru, secondary source). The schedules build on his idea; they are not his.

CAUTIONNonsense syllables make memory measurement "clean"

Murre and Chessa (2023) defend savings as a very "pure" measure. But Bartlett (1932) objected that removing meaning removes what makes everyday memory work. Both have a point.

CAUTIONEbbinghaus found Fechner's book in a second-hand shop in Paris

Universalis and Roediger (1985) say Paris, Wikipedia says London. All sources mention his reading Fechner; where he bought the book is disputed.

FALSE"We forget 70% of what we learn within 24 hours"

The 33.7% after 1 day is relearning effort saved, not the percentage remembered. Subtracting it from 100 and calling it "66–70% forgotten" swaps units. It was also measured on one person's nonsense syllables.

FALSE"Everything we learn is forgotten along exactly this curve"

Meaningful verse was learned far faster (Don Juan, chapter 5). Radvansky, Parra and Doolen (2024) surveyed material from nonsense syllables to novels and found forgetting differs by material and by type of test.

Source: as given in each row; full list in Part 10.

Part 6

What the latest research says

Recent replications and syntheses, with years and sources.

1991 · A German replicationHeller, Mack and Seitz, two people, the same savings method. The curve has the same shape.
1996 · A hundred years of forgettingRubin and Wenzel collect 210 data sets: fast-then-slow is the general rule; the exact function is unsettled.
2006 · 317 experiments on spacingCepeda et al. (Psychological Bulletin): 839 assessments from 317 experiments in 184 articles. The best gap between reviews grows the longer you need to remember.
2015 · A Dutch replicationMurre and Dros (PLOS ONE): about 70 hours of learning over 75 days. The curve replicates, with a small jump from the 24-hour point.
2021 · Spaced self-testingLatimier, Peyre and Ramus (Educational Psychology Review): pooling 29 studies, spaced retrieval practice clearly beats massed retrieval practice (g = 0.74). Expanding schedules were not clearly better than even ones.
2022 · A reviewCarpenter, Pan and Butler (Nature Reviews Psychology): spacing and retrieval practice are powerful but underused, partly because learners hold mistaken beliefs about learning.
2024 · From nonsense syllables to novelsRadvansky, Parra and Doolen (Psychonomic Bulletin & Review) compare forgetting across materials: the more complex the material, the closer to a straight line forgetting is predicted to be.

Source: HistData (Heller et al. 1991); Rubin and Wenzel (1996), Psychological Review 103(4); Cepeda et al. (2006), Psychological Bulletin 132(3), 354–380, abstract on PubMed; Murre and Dros (2015), PLOS ONE 10(7); Latimier et al. (2021), Educational Psychology Review 33, 959–987, abstract (checked via search results); Carpenter et al. (2022), Nature Reviews Psychology 1(9), 496–511; Radvansky et al. (2024), Psychonomic Bulletin & Review 31(6), 2437–2464.

Two curves with the same shape: Ebbinghaus in 1880 and the 2015 replication, both dropping fast then levelling off. The 2015 curve has a small bump at the 24-hour point. Illustrative, not to scale. bump at 24 hours (2015) Ebbinghaus, 1880 illustration of the shape, not to scale
The same shape after 130 years. The 2015 paper concludes the curve has been replicated, and is "not completely smooth".Source: abstract of Murre and Dros (2015), PLOS ONE 10(7): e0120644. The figure is Mark's illustration.
Part 7

Assessment

Scores are Mark's assessment based on the sources in Part 10, not data from a study.

Finding still holds
5/5
Method (for its time)
4/5
Useful today
4/5
Understood correctly online
2/5
Strengths
  • Brought memory into the laboratory for the first time
  • Savings captures even "hidden" memory
  • Replicated more than 100 years later
Weaknesses
  • One person, testing himself
  • Nonsense syllables are far from real lessons
  • The numbers get misquoted as "we forget 70%"

The forgetting curve is real. But the most valuable part of the paper is not the numbers; it is chapter 8: to remember for long, split your reviews over several days.

More: why the "70% forgotten" myth spreads

The number is round, scary, and comes with a scientist's name attached. It sells courses. But it turns "relearning effort saved" into "percentage remembered", two very different things.

The accurate version is less dramatic and more useful: most of the loss happens in the first hour, and a review at the right time pulls it back up.

"More" is Mark's commentary, based on the sources in Parts 4 and 5.

Part 8

Apply it: review before you fully forget

Doable from tomorrow, for anything you need to remember: vocabulary, formulas, lectures.

Four habits: review the same day; review again after about a day; widen the gaps; quiz yourself instead of rereading. 1 · same daya few minutes before bedthe first hour drops fastest 2 · after 1 dayrecall it next daybefore it drops further 3 · widen the gapsdays, then 1–2 weekslonger memory, longer gaps 4 · quiz yourselfclose the book, recalldon't just reread "5 readings in one evening" loses to "5 readings over 5 days".
Each habit rests on a sourced finding. The first hour drops fastest (chapter 7), spread practice beats massed (chapter 8, Cepeda 2006), the best gap grows with how long you need to remember (Cepeda 2006), spaced self-testing is powerful (Latimier 2021).Source: as in Parts 3 and 6. The specific schedule (same day, after 1 day, a few days, 1–2 weeks) is Mark's suggestion, not a schedule a study has proven optimal.
Review the same dayStudied in the morning? Spend 5 minutes recalling it in the evening. That is when the curve is steepest.
Schedule it, don't remember itPut it in your calendar: tomorrow, in 3 days, next week. Review by the calendar, not by mood.
Split it, don't cram itTwo hours to revise? Four 30-minute sessions over four days beat one 2-hour session the night before.
Close the book, then recallEach review, cover the answer and say it first. If you get it, widen the gap; if not, review sooner.
Self-check: how do you review? 5 questions

1. The last time you learned something important, how long before you reviewed it?

2. Is most of your review time crammed into the night before, or spread out?

3. When you review, do you reread or close the book and recall?

4. Try now: write down 5 points from yesterday's lesson without looking. How many did you get?

5. Put 3 reviews of this week's lesson in your calendar. Next week, redo question 4 and compare.

Ebbinghaus and memory techniques are also mentioned briefly in Decoding Super Memory. This piece does not repeat that part.

Part 9

Read next

Ebbinghaus (1885), Memory: A Contribution to Experimental Psychology, 1913 translationFree on psychclassics.yorku.ca. Chapter 7 (the curve) and chapter 8 (spread practice) are short and readable.
Murre and Dros (2015), "Replication and Analysis of Ebbinghaus' Forgetting Curve"PLOS ONE, free to read. Includes a comparison with the original data.
Carpenter, Pan and Butler (2022), "The science of effective learning with spacing and retrieval practice"Nature Reviews Psychology. A readable review of spacing and self-testing.
Radvansky et al. (2022), "A new look at memory retention and forgetting"JEP: Learning, Memory, and Cognition 48(11). A recent review of retention and forgetting.
Papers That Changed How We Learn #02: Miller (1956), the number sevenEbbinghaus could recite a 7-syllable list after one reading. More than 70 years later, Miller asked why seven.
Part 10

Sources

The paper itself and peer-reviewed replications come first. Secondary sources are labelled. The writing environment blocks the original-paper pages and citation databases, so the originals were checked through excerpts of those same pages in search results, not opened in full.

  1. Ebbinghaus, H. (1885). Über das Gedächtnis. Leipzig: Duncker & Humblot. Translation: Memory: A Contribution to Experimental Psychology, trans. Ruger and Bussenius, 1913 (chapters 3, 5, 7, 8). psychclassics.yorku.ca
  2. Murre, J. M. J. and Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLOS ONE 10(7): e0120644. doi.org
  3. Heller, O., Mack, W. and Seitz, J. (1991). Replikation der Ebbinghaus'schen Vergessenskurve mit der Ersparnis-Methode. Zeitschrift für Psychologie 199, 3–18 (cited via HistData).
  4. Friendly, M. R package HistData, dataset Ebbinghaus (data supplied by J. Murre, April 2025). github.com
  5. Murre, J. M. J. and Chessa, A. G. (2023). Why Ebbinghaus' savings method from 1885 is a very "pure" measure of memory performance. Psychonomic Bulletin & Review. doi.org
  6. Rubin, D. C. and Wenzel, A. E. (1996). One hundred years of forgetting. Psychological Review 103(4), 734–760. doi.org
  7. Cepeda, N. J. et al. (2006). Distributed practice in verbal recall tasks. Psychological Bulletin 132(3), 354–380. PubMed
  8. Latimier, A., Peyre, H. and Ramus, F. (2021). A meta-analytic review of the benefit of spacing out retrieval practice episodes on retention. Educational Psychology Review 33, 959–987.
  9. Carpenter, S. K., Pan, S. C. and Butler, A. C. (2022). The science of effective learning with spacing and retrieval practice. Nature Reviews Psychology 1(9), 496–511. doi.org
  10. Radvansky, G. A. et al. (2022). A new look at memory retention and forgetting. JEP: LMC 48(11), 1698–1723; Radvansky, Parra and Doolen (2024). Memory from nonsense syllables to novels. Psychonomic Bulletin & Review 31(6), 2437–2464.
  11. Roediger, H. L. (1985). Remembering Ebbinghaus. Contemporary Psychology 30(7), 519–523; Murdock, B. B. (1985). The contributions of Hermann Ebbinghaus. JEP: LMC 11(3), 469–471; Wickens, T. D. (1998), Psychological Review.
  12. Secondary sources: Encyclopædia Britannica; Wikipedia (Hermann Ebbinghaus, Forgetting curve); Encyclopedia.com; New World Encyclopedia; Encyclopædia Universalis; APS Observer ("Recalling Psychology's Past: The Memory Drum"); SuperMemo and supermemo.guru; J. F. Kihlstrom's notes (Berkeley) quoting Bartlett (1932); db.barbanon.org (chapter 8 quote); idtips.substack.com (the 70% myth).

Citation count: not retrieved (Google Scholar, Semantic Scholar, OpenAlex blocked, checked 15/10/2026).