Memory II: Reconstruction, Interference, and the Limits of Perfect Recall

Two people leave the same meeting with different memories of what was agreed. Neither needs to be dishonest. Each noticed different details, connected them with different expectations, and encountered different information afterward. Remembering involves reconstruction, and reconstruction can preserve meaning while changing particulars.

This second Memory lecture revisits the classic storage model through the experiments behind it, then examines interference, false memories, and Luria’s exceptional mnemonist. It complements the earlier overview of working memory and neural memory systems.

Learning goals and Feynman practice

Explain why remembering differs from replaying a recording, reconstruct the logic of partial-report and distractor experiments, distinguish proactive from retroactive interference, and evaluate what exceptional recall can and cannot establish.

Read a section, close it, and explain it to a beginner. State what the participant did, what changed, and why the result matters. Give a fresh example and one alternative explanation. Repair the step you cannot explain before checking the hidden answers.

1. Camera versus interpretation

Bottom-up information comes from sensory input: the words spoken, the light reaching your eyes, and the sounds in the room. Top-down influences include prior knowledge, expectations, schemas, goals, and later information. These influences interact during perception, encoding, and retrieval; raw input is not first stored in an untouched archive and only interpreted afterward.

Suppose someone says, “We should aim for Friday.” Knowing that Friday is the usual deadline may lead you later to remember, “Friday was confirmed.” The proposal and the confirmation share a plausible meaning, but they are different statements.

A useful analogy is rebuilding a scene from notes and a model of what normally happens. The model helps fill gaps, but a plausible addition can be mistaken for an observed detail. There is no little narrator inside the brain making a deliberate decision to fabricate the scene.

Constructive memory is not necessarily inaccurate. Prior knowledge can organize information and improve understanding. The risk comes when we confuse the source of a remembered element: observation, inference, imagination, or later suggestion.

Teach it back: How could an expectation improve comprehension during a meeting yet make a later quotation less reliable?

2. Three stages: what the classic architecture explains

The Atkinson–Shiffrin model distinguishes sensory registers, short-term storage, and long-term storage, with control processes such as attention and rehearsal. These are theoretical distinctions, not three literal anatomical boxes or a mandatory one-way route for every kind of learning.

Stage	Useful first explanation	Example	Qualification
Sensory memory	Brief persistence of modality-specific information.	A visual array remains partly accessible just after disappearing.	Different modalities and tasks produce different durations.
Short-term memory	Limited temporary retention.	Maintaining a new code before entering it.	Capacity depends on material, rehearsal, knowledge, and interference.
Long-term memory	Retention over extended intervals.	Facts, skills, and events retained for years.	Large capacity does not mean permanent, exact, or literally unlimited storage.

Sperling: ask for a row after the display disappears

Sperling’s 1960 experiments briefly presented letter arrays. In whole report, participants attempted to name everything. In partial report, a tone after the display indicated which row to report. Immediate partial report was substantially better for the selected row, and its advantage declined as the cue was delayed. [1]

The important inference is not that participants could consciously report every letter at once. Because an unpredictable row could be cued after disappearance, more information had briefly been available than whole report could express before it faded.

The outline’s approximate 37% whole-report and 76% partial-report figures illustrate particular conditions, not universal capacity estimates. They also have different denominators: all displayed letters versus the cued subset. To infer total availability from partial report, researchers assume that uncued rows would have been similarly reportable if selected.

Iconic memory lasts a fraction of a second in classic tasks. Attention and report limitations help explain why detailed visual persistence does not become a permanent photographic memory. Sensory memory is not uniformly limited to visual durations; auditory persistence can extend over seconds.

Teach-back check: why must the row cue arrive after the display?

A cue given beforehand would let the participant selectively attend to one row during viewing. A post-display cue instead tests information still accessible after the input has disappeared. The inference still depends on task and cueing assumptions.

3. Short-term capacity and the meaning of a chunk

The classic 7 ± 2 estimate concerns immediate span and meaningful units. It is not a universal limit across short-term tasks. With rehearsal and recoding restricted, estimates often centre on roughly four chunks, although theories and tasks differ. [2]

Read F B I C I A. If you know the acronyms, FBI and CIA organize six letters into two meaningful groups. Chunking uses long-term knowledge to represent structure efficiently. Unknown letters do not become meaningful simply because you draw boxes around them.

Think of packing related objects into familiar kits. You can keep track of the kits more easily, but each kit’s contents depend on what you already know. Chunking expands practical performance without making raw storage limitless.

Peterson and Peterson: why counting backward matters

Participants retained short consonant sequences while performing a distractor task, such as counting backward, before recall. Performance declined substantially over short filled delays. [3]

Counting limits rehearsal, but also introduces other processing. The finding does not establish an exact 18-second expiration time, and later work highlighted proactive interference across trials. The question is why performance declines under those conditions, not merely how long a stopwatch runs.

Teach it back: Why is “recall fell after counting” more defensible than “time alone erased the trace”?

4. Serial position: beginnings, endings, and retrieval conditions

Immediate free recall often produces a U-shaped serial-position curve: beginning and end items outperform middle items.

• Primacy: early items may receive more rehearsal and effective long-term encoding.
• Recency: recent items are especially accessible at immediate recall, classically linked partly to short-term maintenance.

A filled delay commonly reduces immediate recency, while primacy may remain. This is compatible with different contributions to recall, but it does not uniquely prove a two-store explanation. Recency also occurs in longer-timescale tasks, and context and retrieval strategies matter.

“Every 10–30-second filled delay completely eliminates recency” is too strong. Effects depend on presentation, distractor schedule, and recall procedure. Likewise, primacy is not guaranteed after every delay.

Teach-back check: would waiting quietly and counting backward necessarily have the same effect?

No. Quiet waiting permits rehearsal and differs in interference and task demands. Compare the delay conditions rather than attribute every difference to elapsed time alone.

5. Forgetting: the passage of time versus competing information

Ebbinghaus documented substantial early forgetting followed by a slower decline under his learning conditions. A forgetting curve describes performance over time. It does not, by itself, show why forgetting occurs or establish one universal mathematical function.

Proactive interference: older learning disrupts newer learning or retrieval. Your old password repeatedly comes to mind when you try to use its replacement.

Retroactive interference: later learning disrupts access to earlier learning. Repeatedly using the replacement makes the old password harder to retrieve when you need it for another system.

The terms describe the direction of influence relative to the target memory, not an inevitable deletion of stored information. Interference is important, but encoding quality, changing cues, retrieval competition, and changes in representations also matter. The evidence does not justify saying that passive decay never contributes.

Wickens: release from proactive interference

Imagine several successive trials containing fruit words. As related lists accumulate, earlier items compete with the current list. On a later trial, the category switches to professions. Recall can improve despite continued testing: release from proactive interference.

Wickens used such changes to investigate dimensions along which words are encoded. [4] If changing meaning releases interference more than changing an irrelevant visual feature, that supports semantic organization under the tested conditions.

A category switch does not guarantee a return to perfect performance. To interpret it, compare with participants who continue in the original category and keep timing and difficulty comparable. The fruit-to-profession sequence here is a teaching example rather than a reproduced original dataset.

Teach it back: Why would improvement on a later trial challenge the claim that fatigue or elapsed time alone explains the earlier decline?

Jenkins and Dallenbach: sleep versus waking

Their 1924 study found better retention following sleep than comparable waking intervals. [5] This supports examining what happens during the interval rather than treating time as the only cause of forgetting.

The historical comparison does not isolate interference from every other difference between sleep and wake. Sleep can also actively support consolidation; later experiments demonstrate increased resistance to subsequent interference. [6] “Sleep prevents waking experiences from overwriting memory” is therefore only a partial account, and “overwriting” itself is a metaphor.

Teach-back check: does better recall after sleep prove that no active memory processing happened during sleep?

No. Reduced waking interference and active consolidation can both contribute. To distinguish them, manipulate interference and examine later resistance while controlling relevant timing and learning differences.

6. Gist and false memories: meaning can masquerade as an event

Gist is an overall meaning or theme. Verbatim information concerns specific details. Both can be retained, and their relative accessibility changes with conditions; memory does not always discard detail and keep only gist.

In the Deese–Roediger–McDermott, or DRM, paradigm, people study associated words and may later recall or recognize a related word that was not presented. Roediger and McDermott’s experiments demonstrated substantial false remembering using such lists. [7]

For example, candy, sugar, honey, and syrup can activate the unpresented associate “sweet.” At retrieval, familiarity or thematic fit may be attributed to actual presentation. Associative activation, gist, and source-monitoring processes offer complementary ways to investigate the error.

This four-word illustration is not a standardized DRM experiment. The outcome depends on list construction, instructions, delay, and awareness of the demonstration. Do not assume most participants will add a specific word under every condition. A laboratory word-list error also does not establish that an unrelated autobiographical memory is false.

Feynman exercise: separate observation from inference

Write two columns after learning a concept: what the source actually said and what I inferred. Then explain the difference aloud. For the meeting example, put “aim for Friday” in the first column and “Friday is confirmed” in the second. Understanding both makes it easier to check their source later.

Teach it back: Why can explaining the general theme accurately coexist with remembering an unpresented word?

7. Luria’s mnemonist: extraordinary recall is not all-purpose understanding

A. R. Luria’s The Mind of a Mnemonist describes long-term observations of S., Solomon Shereshevsky, who showed remarkable retention and reported vivid synesthetic experiences. Sounds and other stimuli evoked additional sensory qualities, and elaborate imagery and mnemonic organization helped his recall. [8]

Luria also described difficulties with abstraction, figurative language, and managing intrusive detail. A metaphor could evoke concrete images that interfered with grasping its intended meaning. The case illustrates a possible tension between retaining particular detail and efficiently extracting general structure.

It does not establish literally limitless memory, universal verbatim accuracy, or that synesthesia alone caused the ability. Not finding a limit in the tested tasks is different from proving infinity. Nor does one exceptional case prove that strong memory generally impairs reasoning, that all synesthetes are mnemonists, or that S. could extract no gist at all.

A useful analogy is an archive containing unusually rich records but requiring effort to identify the pattern across them. Detail and generalization are different achievements, even though they can support each other.

Teach-back check: why is one exceptional case informative but insufficient for a general causal claim?

It can reveal an unusual combination of abilities and suggest mechanisms. Without appropriate comparisons and interventions, it cannot isolate which feature caused the others or establish that the same relationship holds across people.

8. Worked comparison: what does each experiment establish?

Finding	Supported inference	Stronger claim to avoid
Immediate partial report beats whole report.	More information is briefly accessible than can be fully reported.	Every detail is consciously stored as a durable photograph.
Recall falls across a filled delay.	Retention suffers under that delay and distractor procedure.	All short-term memories expire at exactly 18 seconds.
A category switch improves recall.	Category-sensitive competition contributes to performance.	All earlier information was erased and then restored.
Sleep exceeds waking retention.	Interval conditions matter.	Sleep benefits are exclusively passive.
A related, absent word is remembered.	Meaning and source attribution can produce an error.	Every vivid recollection is unreliable.
S. retains exceptional detail.	Memory abilities can be unusually configured.	Unlimited recall or a universal memory–intelligence trade-off.

Retrieval workshop and spaced review

Without looking, explain the three-stage model, the post-display cue, chunking, primacy and recency, and both directions of interference. Then invent a category-switch experiment and explain why its control group matters.

Repair: “Because memory is constructive, it cannot provide evidence”

Constructive processing can preserve accurate information as well as introduce errors. Evaluate the conditions of encoding and retrieval, separate details from inferences, and seek corroboration where precision matters. Reconstruction is not the same as arbitrary invention.

Apply: distinguish a forgotten memory from a failed retrieval attempt

Try a relevant cue or another retention measure, such as recognition or relearning. Improved performance can show that free recall did not exhaust available information. It still does not establish a fully intact hidden recording.

Review tomorrow and in a week. Score each explanation: 0 for a label, 1 for a definition, 2 for a mechanism plus example and limitation. Repair the weakest explanation before adding more terminology.

One-minute explanation: Memory preserves selected information through several interacting processes. Brief sensory persistence exceeds what we can report, temporary retention is limited, and long-term knowledge helps organize new material. Forgetting reflects more than time alone: competition and retrieval conditions matter. Meaning helps us generalize but can also be confused with observed detail. Exceptional recall shows why remembering particulars and understanding principles must be assessed separately.

Sources and further reading

The supplied numbered citations did not include a bibliography. These references support the experiments and case account; the meeting, password, and category-switch examples are teaching aids.

1. Sperling, G. (1960). The information available in brief visual presentations.
2. Cowan, N. (2001). The magical number 4 in short-term memory.
3. Peterson, L. R., and Peterson, M. J. (1959). Short-term retention of individual verbal items.
4. Wickens, D. D. (1970). Encoding categories of words: an empirical approach to meaning.
5. Jenkins, J. G., and Dallenbach, K. M. (1924). Obliviscence during sleep and waking.
6. The sleeping brain’s influence on verbal memory: boosting resistance to interference.
7. Roediger, H. L., and McDermott, K. B. (1995). Creating false memories: remembering words not presented in lists.
8. Luria, A. R. The Mind of a Mnemonist: A Little Book about a Vast Memory. Harvard University Press edition and preview.