Forgetting — What Fades and What We Just Can't Get Out

Yesterday's Lunch and Last Tuesday's Lunch

Try to recall what you had for lunch yesterday. It usually comes to mind right away. You might even remember it was kimchi stew, who you ate with, and that the broth was a little salty.

How about last Tuesday's lunch? Many people get stuck here. You definitely ate something, but what it was just won't come up.

Then a coworker sitting next to you chimes in: "We went to the kalguksu place in front of the office because it was raining that day." The moment you hear those words, the wet umbrella, the steaming bowl, and the window seat all rush back at once. Where did the memory that wouldn't come up until a minute ago come from?

Like the "The Three Steps of Memory — Putting It In, Keeping It, Taking It Out" piece, this article compares memory to a library. A memory to keep for a long time is a book shelved on the rack, and a memory you've brought to mind right now is a book taken out onto the desk. A book that won't come to mind falls into three cases. A book that never made it to the shelf in the first place, a book that was shelved and then truly disappeared, and a book that is on the shelf but cannot be found. The story of books that didn't make it to the shelf is covered by "The Three Steps of Memory", and this article looks at the latter two cases after a book goes onto the shelf. Tuesday's lunch, which returned at a single word about the kalguksu restaurant, was the last case. The lunch story is an example created by this article to help understanding.

Fast at First, Slow Later

Time plays a major role in why yesterday's lunch is easy to remember while last week's lunch is not. What we learn drains away quickly during the first little while after we learn it (PMID 36880338). From then on, the speed at which it drains slows down. If you have two memories that are equally clear right now, the older one will blur more slowly, a law that is over 100 years old (PMID 15482065).

The forgetting curve is a line that plots how much memory remains as time passes after learning. It drops steeply at first, and becomes more gradual as time goes on. This curve was first drawn by the psychologist Ebbinghaus in the 1880s (DOI: 10.1371/journal.pone.0120644). Because this curve was drawn using data from memorizing lists, it does not mean that things learned with understanding fade at the exact same speed.

Ebbinghaus measured how much memory remained in terms of savings (DOI: 10.1371/journal.pone.0120644). Savings is the effort that takes less than the first time when you re-memorize something you memorized once before. If a list that took 10 minutes to memorize the first time takes 6 minutes to memorize again a day later, the reduced 4 minutes is the savings. These 10 minutes and 6 minutes are numbers made up for the sake of explanation. Measured this way, even on a day when you cannot recall a single item from the list, if the re-memorizing time is shorter than the first time, you can count that something remained.

A study that redrew this curve yielded similar results. The results from one person spending 70 hours memorizing lists and re-memorizing them at six time points ranging from 20 minutes later to 31 days later were similar to Ebbinghaus's old data (DOI: 10.1371/journal.pone.0120644). However, the curve was not completely smooth, and the authors consider it likely that the amount remaining jumps up once from the smooth curve starting from the data on day one (DOI: 10.1371/journal.pone.0120644). The point that it is data from a single person is the same as the old experiment.

There are also objections that this curve is not a law that fits all memories. The claim is that some memories remain entirely or disappear entirely, resulting in a staircase shape instead of a smooth curve. This objection is a theoretical manuscript posted to an open repository in 2026, which has not undergone peer review. It is a claim built on logic without data, and experiments to sort it out are still at the stage of proposing designs (DOI: 10.5281/zenodo.18945416).

In a curve drawn from an experiment where one person memorized lists, forgetting is fastest during the first few hours after learning and grows slower as time passes. Lunch from last Tuesday is a memory that has already passed well beyond that steep section.

What Was Erased and What Could Not Be Pulled Out

A colleague asks again, "What was the name of that kalguksu restaurant?" You know the sign was green, and you know the name starts with "ㅁ", but the name itself does not come out. Anyone knows that feeling of it being right on the tip of your tongue yet not coming out.

The tip-of-the-tongue phenomenon is a state where a known word hovers at the tip of the tongue and briefly fails to come to mind. It is also called the tongue phenomenon, and it comes with a strong feeling that it will be recalled soon (PMID 41751168). The tip-of-the-tongue phenomenon is experienced by almost everyone, visits about once a week, and frequently occurs with proper nouns like people's names (DOI: 10.1037/0033-2909.109.2.204). Amid that, the first letter often comes to mind, other similar words pop out instead, and about half are resolved on the spot (DOI: 10.1037/0033-2909.109.2.204). However, the feeling that it will come to mind soon is a guess created by various clues like the familiarity of the recalled partial information or cue, so we cannot conclude that the word is in memory based on the feeling alone (PMID 21264637).

If the first letter comes out correctly, the book is likely on the shelf, and the block is a matter of finding and pulling it out. Retrieval is the act of finding a memory on the shelf and pulling it back onto the desk. Retrieval failure is the inability to pull out a memory at that moment even though it remains. The tip-of-the-tongue phenomenon is retrieval failure shown on the outside, and even while a known word cannot be pulled out, knowledge of that word's meaning remains intact (PMID 41751168, PMID 21264637).

Retrieval failure can sometimes be resolved when a suitable cue arrives. A cue is a clue for finding a memory, or in library terms, a spine label or index. When people found a smell familiar but could not name it, giving a hint often brought the name out (DOI: 10.1037/0278-7393.3.1.52). One major variable governing forgetting is how much the cue and scene at the time of recall resemble the scene at the time of learning (PMID 33794376). The "rain" that summoned Tuesday's lunch was something contained in that day's scene. How great the help is from reviving the scene at the time of storage is covered separately in "The Three Steps of Memory", and the result seen in that piece was that it tended not to be very large.

Then, are forgotten memories erased, or could they not be pulled out? The answer put forward by one researcher is "mostly they could not be pulled out." After examining various branches of forgetting, the author of a review article collecting and organizing multiple studies concludes by viewing most forgetting as retrieval failure rather than erasure that cannot be undone (PMID 33794376). The reason it is "mostly" and not "all" is because explanations of the side where it truly disappears also remain. Fading away naturally over time is one of the old explanations of forgetting (PMID 33794376). In studies on fruit flies and mammalian animals, evidence has also emerged that there is a process where the brain actively deletes memories it has already made (PMID 29467630, PMID 24767478). However, this is the result of animal experiments. The author of the previous review considers it logically difficult to say that a memory was "erased" in the case of common forgetting (PMID 33794376). This is because the mere fact that it cannot be pulled out right now cannot distinguish whether the memory has disappeared or whether a suitable cue has not yet been met.

Quite a few memories that do not come to mind remain on the shelf and come back out when a suitable cue arrives.

Similar Books Get Mixed Up

A weekday lunch happens about 20 times a month. You usually eat at a similar time, with similar people, in a similar neighborhood. In library terms, it looks like 20 books with nearly identical spines crammed onto a single shelf.

Interference is when similar, other memories push in and make it hard to retrieve the memory you are trying to find. Information that resembles the target information yet is distinct ranks among the primary causes of forgetting (PMID 33794376). Acquiring new memories and thereby forgetting other, related memories stands as one of the most fundamental phenomena in memory research (PMID 16890518).

The direction of disruption goes two ways. What is experienced later can overwrite earlier memories, and what is experienced earlier can block later memories (DOI: 10.1037/0033-295x.95.4.434). Translated to lunches, this week's lunches overwrite last Tuesday. Conversely, when trying to recall this week's Monday lunch, the noodle shop you went to last Monday might pop up instead.

How large the disruption is depends on two degrees of similarity. One is how much the memory you are looking for resembles the clue you currently hold in your hand, and the other is how much the intruding memory resembles that same clue. How long ago the target memory and the intruding memory were each formed also plays a role (PMID 33794376). Applied to lunch, it works like this. People usually do not separately count the days while eating lunch, so the label "last Tuesday" is not attached separately to any of the 20 books. Because the target book and the remaining intruding books resemble this clue to the same degree, this clue cannot single out one book. On the other hand, "lunch on a rainy day" is attached to only one book that day, resembling the book you are looking for while not resembling the books that are intruding. Putting these two similarities together in this article, the more a clue is contained solely within that specific memory, the easier it is to pick out the book you are looking for from among similar books.

Pull One Out and the Neighbor Gets Pushed Aside

Imagine chatting with colleagues over lunch about last week, and only the new pork cutlet restaurant on Monday and the team dinner on Wednesday keep coming up. The more you talk about those two days, the clearer they become. Yet if you try to recall Tuesday right then and there, Tuesday—tied to the same "last week's lunch"—can feel even further away.

Retrieval-induced forgetting is a phenomenon where pulling out a certain memory makes other memories tied to the same cue harder to bring to mind later. In one experiment, people memorized eight words from categories like "fruit." Next, for some categories, they repeatedly practiced pulling out only half of those category words by looking at the first letter, such as "fruit–or○○." On a later test, the words that belonged to the practiced categories but were not practiced themselves were harder to recall, and this effect lasted for at least 20 minutes (DOI: 10.1037/0278-7393.20.5.1063). This penalty mainly appeared in frequently used, common words, while rare words suffered almost no penalty (DOI: 10.1037/0278-7393.20.5.1063).

The memories that were pulled out themselves stuck around better. In experiments memorizing colors and shapes, practiced items were remembered better, while unpracticed items in the same batch suffered a penalty (DOI: 10.1037//0278-7393.25.6.1403). This penalty did not happen simply by showing the items one more time; it only occurred when they were actively pulled out (DOI: 10.1037//0278-7393.25.6.1403). Here is where the library analogy breaks down once. Books do not become sturdier just because you pull them out, but memories stick around better when you pull them out and look at them. Why studying by pulling things out makes them stick is covered in "Retrieval Practice — The More You Pull It Out, the More It Stays".

The effect of pushing aside neighboring memories did not always last long. In one experiment, the effect appeared when tested immediately after practicing retrieval, but did not appear when tested a day after practice (DOI: 10.1111/1467-9280.00325). When the order was reversed so they memorized first and practiced a day later, the effect diminished but did not disappear (DOI: 10.1111/1467-9280.00325). In word list experiments, people who thought about words from the same category in sequence while memorizing experienced less of the effect (DOI: 10.1037/0278-7393.25.3.608).

This phenomenon also happens in conversations. Even people who merely listened to someone else selectively recall things forgot more of the unmentioned, related details themselves (DOI: 10.1111/j.1467-9280.2007.01967.x). Why this happens is still debated. An explanation that the brain suppresses memories competing with the one being retrieved clashes with an explanation that it can be accounted for solely by competition between memories without suppression. An analysis gathering multiple studies generally supported the suppression side, but there was also counter-evidence, and results varied depending on how it was measured (PMID 25180807).

There Is a Use for Forgetting, Too

Let's say you are trying to remember where you made lunch plans today. If all 20 meeting spots from the past month were as vivid as yesterday, picking out today's spot would actually become harder.

Adaptive forgetting is a kind of forgetting where memories that have served their purpose blur, making it easier to find the memories you need right now. There is a theory that for information that needs to be rewritten often, the old value must blur so the new value is less hindered (DOI: 10.1111/1467-9280.00405). In an experiment where the task to be done changed every few seconds, responses gradually slowed and errors increased as time passed after a change, and when the interval between changes grew longer, the rate of this decline also slowed (DOI: 10.1111/1467-9280.00405). The authors read this to mean that the rate of blurring is tuned to the rate of updating. However, this is a lab task on a scale of a few seconds.

The author of a review dividing the flaws of memory into seven types argues that even though those flaws look like design defects, they are byproducts produced by useful properties of memory in other ways (DOI: 10.1037/0003-066x.54.3.182). The beneficial outcomes brought by forgetting are often overlooked (PMID 33794376). To use a library analogy, books that are not in use need to be pushed to the inner shelves so that the books you use often can be reached.

A Retrieved Memory Can Change

There is one more place where the library analogy falls short. Books stay the same even when you take them out and put them back, but memories may not. The act of recalling is closer to gathering scattered pages and binding them together on the spot—just as we saw in the primer "The Three Steps of Memory"—than it is to taking out and reading a book as it is. There is a hypothesis that even a once-consolidated memory enters a temporarily unstable state the moment it is retrieved and can change at that time, and this hypothesis still remains controversial (PMID 19073438). This phenomenon was first reported in various animal species and in human motor skill memories, and for human verbal memories, an experiment having people memorize paired syllables provided evidence (DOI: 10.1101/lm.486107).

When you rebind them, things that were not there can sometimes get mixed in. A word list experiment demonstrating this plays words surrounding a theme while leaving out the theme word itself. For example, it plays "bed, rest, dream, blanket, yawn" but leaves out "sleep". When 111 such experimental results were gathered, people believed the unheard theme word was in the list and recalled it at a rate of roughly 37%. The rate of recalling actually heard words was 57%, so that is about two-thirds of it (PMID 15002841). When told in advance not to guess, the rate of recalling unheard theme words decreased. Those who received such instructions were 32%, and those who did not were 43%. Even so, it did not disappear (PMID 15002841).

If there is no other evidence to verify it, there is no test that can reliably distinguish between true and false memories (PMID 7960290). Even a vividly recalled memory is no guarantee that it remains unchanged.

When You Need to Find Last Tuesday's Lunch

Next Monday, the accountant asks, "That lunch receipt from last Tuesday, which restaurant was it?" Let's trace what we saw in this article, step by step.

First, we don't stop at "I can't remember." Quite a few of the memories that won't surface are still sitting on the bookshelf. Because the label "last Tuesday" isn't specifically attached to any of the twenty lunches, we look for something that happened only on that day. Things like the weather, the person we went with, or the meeting that morning. When the cue at the time of recall resembles the scene at the time of experience, we generally remember a little better, and cues unique to that memory are especially helpful.

When the restaurant's name is on the tip of our tongue, we recall that the first letter or a similar name often comes to mind first, and about half of those instances get resolved right then and there.

If we keep repeating only the stories of Monday's pork cutlet restaurant and Wednesday's team dinner, Tuesday might temporarily drift further away right then and there. In experiments, this effect sometimes did not appear in tests taken one day after practice.

Even when we finally remember the noodle shop, before handing in the receipt, we take a look at the card statement. There is no test that separates truth from falsehood using only memory without evidence to confirm it.

Translating this to studying, it looks like this. Contents that resemble each other, like two formulas with similar appearances, can interfere with each other. In word list experiments, people who thought about items in the same group back-to-back experienced less of the effect where neighboring memories are pushed out. The method of binding where they are the same and where they are different onto a single sheet rather than memorizing two formulas separately is a translation of this result, and the results of testing this method directly in a classroom are not among the evidence for this article. Because what is learned fades the fastest at first, the next question becomes when to set the time for reviewing it.

Remaining Questions, and Where to Look Next

It has not yet been concluded how much of the memories that entered the bookshelf actually disappeared and how much is on the bookshelf but cannot be found.

How memories are initially put in and kept is covered in "The Three Steps of Memory — Putting It In, Keeping It, Taking It Out". Why the act of pulling them out leaves memories behind continues in "Retrieval Practice — The More You Pull It Out, the More It Stays", and when to set the time for looking at them again continues in "The Spacing Effect — Spread It Out and It Lasts". To take a single sweep over which study methods actually work on top of this foundation, you can go to "The Science of Studying: What Actually Works".

Source Papers

• "Replication and Analysis of Ebbinghaus' Forgetting Curve" (2015) — DOI: 10.1371/journal.pone.0120644 — Replication of Ebbinghaus's savings method where a person memorized lists for 70 hours and re-memorized them 20 minutes to 31 days later, showing results similar to the original data and the possibility of bouncing upward starting on day one.
• "Spaced Effect Learning and Blunting the Forgetfulness Curve." (2022) — PMID 36880338 — A medical education review noting that material encountered in lectures and study is forgotten rapidly over the following few days, with spaced repetition used as the solution.
• "On Common Ground: Jost's (1897) law of forgetting and Ribot's (1881) law of retrograde amnesia." (2004) — PMID 15482065 — A review covering the implication of Jost's law that between two memories of equal strength but different ages, the older one fades more slowly.
• "The Step Function: Against the Universality of Graded Forgetting" (2026) — DOI: 10.5281/zenodo.18945416 — (A preprint not peer-reviewed) A theoretical manuscript proposing that the forgetting curve is not a universal law and that memory structures may exist in which memories remain or disappear in a step-like fashion, proposing a design for verification experiments.
• "The Tip-of-the-Tongue Phenomenon: Cognitive, Neural, and Neurochemical Perspectives." (2026) — PMID 41751168 — A review describing the tip-of-the-tongue phenomenon as a state where a known word cannot be retrieved temporarily while meaning knowledge is preserved, accompanied by a strong feeling that it will be recalled soon.
• "Tip-of-the-tongue (TOT) states: retrieval, behavior, and experience." (2011) — PMID 21264637 — A review defining tip-of-the-tongue states as the feeling that follows when an item to be retrieved cannot be temporarily accessed, explaining retrieval failure and that experience together.
• "A review of the tip-of-the-tongue experience." (1991) — DOI: 10.1037/0033-2909.109.2.204 — An overview noting that the tip-of-the-tongue experience is nearly universal, occurs about once a week, is frequent with proper nouns, prompts the first letter or related words to come to mind, and is resolved on the spot about half the time.
• "Associations to odors: Interference, mnemonics, and verbal labeling." (1977) — DOI: 10.1037/0278-7393.3.1.52 — An experiment showing that in a state where familiar smells could not be named, hints frequently brought the correct names to mind.
• "Failures of memory and the fate of forgotten memories." (2021) — PMID 33794376 — A review categorizing the four branches of forgetting (interference, decay, short-term memory displacement, and insufficient cues), concluding that most forgetting is retrieval failure rather than erasure, and covering the benefits and moderating variables of forgetting.
• "Neural, Cellular and Molecular Mechanisms of Active Forgetting." (2018) — PMID 29467630 — A review summarizing the neural processes that actively erase memories in fruit flies and mammals.
• "Active forgetting of olfactory memories in Drosophila." (2014) — PMID 24767478 — A review organizing evidence from olfactory memory research in fruit flies showing that forgetting is an active process regulated biologically.
• "Declarative memory: sleep protects new memories from interference." (2006) — PMID 16890518 — A commentary noting that interference—the phenomenon where a new memory causes other related memories to be forgotten—is one of the most fundamental phenomena in memory research.
• "A model for interference and forgetting." (1988) — DOI: 10.1037/0033-295x.95.4.434 — A memory search model explaining classical interference results, such as the phenomenon where things learned later interfere with earlier things, and earlier things interfere with later ones.
• "Remembering can cause forgetting: Retrieval dynamics in long-term memory." (1994) — DOI: 10.1037/0278-7393.20.5.1063 — An experiment demonstrating retrieval-induced forgetting, where repeatedly recalling some category words makes unpracticed words from the same category harder to recall for 20 minutes or longer.
• "Retrieval-induced forgetting in episodic memory." (1999) — DOI: 10.1037//0278-7393.25.6.1403 — An experiment showing that practiced items improve while unpracticed items in the same set suffer, and that the damage stems from retrieval rather than re-exposure.
• "Gone but Not Forgotten: The Transient Nature of Retrieval-Induced Forgetting" (2001) — DOI: 10.1111/1467-9280.00325 — An experiment where retrieval-induced forgetting did not appear when the interval between practice and the final test was 24 hours, and was reduced but not eliminated when practice occurred a day after memorization.
• "Integration as a general boundary condition on retrieval-induced forgetting." (1999) — DOI: 10.1037/0278-7393.25.3.608 — Three experiments showing that retrieval-induced forgetting was reduced when words from the same category were memorized by linking them to one another.
• "Silence Is Not Golden" (2007) — DOI: 10.1111/j.1467-9280.2007.01967.x — An experiment showing that people who merely listened to someone else selectively recall items forgot even more of the unmentioned related information.
• "Forgetting as a consequence of retrieval: a meta-analytic review of retrieval-induced forgetting." (2014) — PMID 25180807 — The first large-scale meta-analysis of retrieval-induced forgetting, where results generally supported the inhibition explanation, though contradictory evidence also existed and varied depending on the measurement method.
• "Forgetting to Remember: The Functional Relationship of Decay and Interference" (2002) — DOI: 10.1111/1467-9280.00405 — A theory and task-switching experiment proposing that values requiring frequent updating fade to prevent interference from subsequent values, with that speed matched to the frequency of updating.
• "The seven sins of memory: Insights from psychology and cognitive neuroscience." (1999) — DOI: 10.1037/0003-066x.54.3.182 — An overview arguing that the seven flaws of memory are not design defects but byproducts of adaptive traits.
• "Reconsolidation and the fate of consolidated memories." (2008) — PMID 19073438 — A review addressing the hypothesis that even consolidated memories become unstable again when retrieved and undergo reconsolidation, along with the controversies surrounding it.
• "Reconsolidation of declarative memory in humans" (2007) — DOI: 10.1101/lm.486107 — An experiment presenting evidence that reconsolidation also occurs in human declarative memory during paired-syllable learning.
• "False recall with the drmrs ("drummers") procedure: a quantitative summary and review." (2003) — PMID 15002841 — Across 111 estimates, the average false recall of unpresented theme words was .374, correct recall was .572, and under instructions not to guess, it was .318 versus .428.
• "Hypnosis, delayed recall, and the principles of memory." (1994) — PMID 7960290 — A review stating that without independent verification, there is no reliable test to distinguish true memories from false ones.