Modifying memory: selectively enhancing and updating personal memories for a museum tour by reactivating them

Psychological Science 24(4): 537–543, April 2013 (online 13 February 2013), DOI 10.1177/0956797612457377. Read as the NIH author manuscript (PMC3805141, NIHMS517124; not the publisher's version). Harvard University. Empirical, two experiments, peer-reviewed. Provenance: papers/carry_on/stjacques2013_reactivation.provenance.json.

What was read

• Read: all 303 lines of the text derived from the PMC XML. That covers the abstract, introduction, Studies 1 and 2 (method and results), discussion, acknowledgements, funding, the conflict-of-interest statement, all 33 references, the captions of Figures 1–2, and Table 1 with its note.
• Not read:
  • Figures 1–2 are images; their captions were read.
    • Figure 1 shows the trial sequence.
    • Figure 2 holds the difference scores by condition and, for Study 2, by high- and low-reliving group. For those groups the text gives only test statistics; Table 1 gives the rates for the whole samples.
  • There is no supplement (PMC: has-supplement no).

Question

When a person is reminded of a real, recent experience, does the quality of the reminder decide two things?

• How much the memory is strengthened.
• How much new, related information shown just after the reminder is taken into the memory.

The authors predicted that reactivation would do both, raising later true and false recognition. They also predicted that a reminder matching the original experience would do more than a mismatching one.

Method

• Encoding (both studies).
  • Participants took a self-guided tour of the adjoining Harvard Natural History and Peabody museums, following a tour booklet.
  • They wore a ViconRevue camera that "automatically takes photos every 15 s".
  • People who had visited the museums before were excluded.
  • There were two versions of the tour, with the same events but different stops within each event. The other version's stops served as lures; the versions were counterbalanced.
  • Each participant's photos were checked. An event was excluded if the camera caught an alternate stop.
• Session 2, reactivation, 48 h after the tour.
  • For each reactivated event, participants saw a "movie" of photos of that event's stops.
  • After 0.5 s came a novel photo of an alternate stop at the same exhibit, one the participant had not visited.
  • Participants judged, yes or no, whether that photo was related to the event.
• Session 3, test, 48 h after reactivation. Yes/no recognition with confidence ratings.
• Study 1.
  • Sample: N = 42 (27 women, 15 men; mean age 21.11 years, SD 2.87). Two more were excluded.
  • Tour: 32 events of six stops each; the last two stops of each event differed between versions.
  • Movie: six photos at 1.25 s each.
  • Conditions:
    • match: stops in the order experienced;
    • mismatch: the last four stops shown in a changed order;
    • three quarters of events were reactivated, the rest served as baseline.
  • Test:
    • pairs of photos: two targets, or a target and a lure;
    • answer "yes" only if both stops "had been experienced together during an event at the museum";
    • confidence on a 4-point scale.
• Study 2.
  • Sample: N = 41 (23 women, 18 men; mean age 21.44 years, SD 2.22). One more was excluded.
  • Tour: 53 events of four stops each.
  • Conditions:
    • match: the movie showed the participant's own photos;
    • mismatch: photos from a control set with the angle and/or height changed;
    • the novel photo had a typical perspective in match trials and an atypical one in mismatch trials;
    • half of the events were reactivated.
  • After each trial, participants rated their confidence in the relatedness judgement and their sense of reliving while watching the movie (both 1–5).
  • Test:
    • single photos;
    • "yes" only to stops "experienced during their museum tour";
    • confidence 1–5;
    • separate baselines for own-perspective and altered-perspective photos.
• Analysis.
  • The measures are difference scores: the hit or false-alarm rate in a reactivation condition minus that in its baseline.
  • Repeated-measures ANOVAs.
  • Study 2 adds a median split on each person's reliving difference (match minus mismatch).

Results

Session 2, reactivation.

• Study 1.
  • Reaction times did not differ: match 3.77 s (SD 0.37 s), mismatch 3.87 s (SD 0.38 s).
  • Correct relatedness judgements did not differ either: .85 against .84 (SEM .03 each).
• Study 2.
  • Reaction times: 2.61 s (SD 0.66 s) against 2.70 s (SD .73 s), no difference. Correct judgements: .85 against .84 (SEM .03 each), no difference.
  • Confidence was higher after a match: 4.01 (SEM 0.08) against 3.66 (SEM 0.08), t(39) = 4.32, p < .0001.
  • Reliving ratings were faster after a match, 1.38 s (SD 0.52 s) against 1.54 s (SD 0.46 s), t(40) = 4.08, p < .0001.
  • Reliving itself was higher: 3.79 (SEM 0.07) against 3.27 (SEM 0.08), t(40) = 5.82, p < .0001.
  • The authors take this as showing that the perspective manipulation worked.

Session 3, recognition (Table 1; mean proportions, SEM in brackets):

• Study 1.
  • Main effect of memory type: F(1, 39) = 25.16, p < .0001, ηp2 = .39. "prior exposure had a greater influence on the proportion of false alarms than on the proportion of hits (relative to baseline)".
  • Main effect of reactivation condition: F(1, 41) = 8.59, p < .01, ηp2 = .17. Hits and false alarms both rose more after a match than after a mismatch.
  • There was no interaction.
  • Confidence and reaction times did not differ between the two reactivation conditions.
• Study 2.
  • Confidence. Confidence difference scores (reactivated minus baseline) were higher for false alarms (M = 0.52, SEM = 0.17) than for hits (M = −0.01, SEM = 0.06): F(1, 40) = 8.42, p < .01, ηp2 = .17. Only the difference between the two is tested, not each against zero.
  • Reliving moderates the reactivation effect. Reactivation condition × reliving: F(1, 39) = 9.33, p < .005, ηp2 = .19.
    • Match and mismatch differed only in the high-reliving group: F(1, 20) = 9.12, p < .01, ηp2 = .31. There, reactivation "increased both hits and alarms significantly more" after a match.
  • Main effect of memory type: F(1, 39) = 25.16, p < .0001, ηp2 = .39.
  • The reactivation condition alone was only marginal (p = .09).
  • Both studies pooled: reactivation condition F(1, 81) = 8.40, p < .005, ηp2 = .09, with no interaction with study.
• Our computations from Table 1 (from the group means, so exact for means but untested).

  • Difference scores (reactivated minus baseline):

    	Hits	False alarms
    Study 1, match	+.12	+.26
    Study 1, mismatch	+.06	+.21
    Study 2, match	+.08	+.23
    Study 2, mismatch	+.05	+.16

  • Hits minus false alarms:

    • Study 1: .13 after a match, .12 after a mismatch, .27 at baseline;
    • Study 2: .28 against .43 for match versus its baseline, .14 against .25 for mismatch versus its baseline.

  • So on these means, hits minus false alarms was lower for reactivated events in every comparison: people told experienced stops from unvisited ones less well. This is untested, and the paper reports no measure of discrimination or bias.

Discussion.

• The authors' claim: "Here we have shown, for the first time, that manipulating properties of reactivation selectively influences personal memories by both enhancing and distorting memory via updating."
• Source monitoring.
  • The false alarms are source errors by definition: people claimed that something seen after the tour happened during it.
  • The authors argue that simple source confusion is controlled, because post-event photos were shown in both reactivation conditions.
  • They accept that post-event photos may have been harder to tell from targets after a matched cue. They see no conflict between that and their reactivation account, and call for further work on how the two relate.
• Boundary conditions. Reactivation "may not affect all memories equally in all contexts". It may depend on the strength of the memory and on the nature of the new information.
• Everyday reminders. "Reminders of past experiences occur frequently in daily life", which lets memories be strengthened and updated "so that memories continue to remain relevant in the future".

Limits

• Narrow sample and short delays.
  • Two small samples (42 and 41) of young adults, mean age about 21, in one pair of museums.
  • The delays are 48 h and 48 h, so the paper says nothing about months or years.
• Recognition only. Photo recognition, with no free recall and no narrative.
• Mechanism undecided (our assessment).
  • The introduction frames the study with reconsolidation. Nothing in the design separates reconsolidation-based updating from source confusion or familiarity: there is no blocker, no delay manipulation and no test of lability.
  • The authors accept that the effect can be described as harder source monitoring after a matched cue ("We do not dispute this characterization").
• "Better memory" means more hits (our assessment).
  • False alarms rose more than hits, and hits minus false alarms fell (see Results).
  • A more liberal criterion for strongly reactivated events would also raise hits and false alarms together, with no interaction. The paper does not separate discrimination from bias.
• Re-exposure.
  • In Study 1 a photo of each stop was selected "to use in the later sessions", and the event movies showed all six stops. So reactivated targets had been seen again in Session 2 and baseline targets had not (our reading).
  • The hit gain over baseline therefore includes plain re-exposure. The match–mismatch contrast controls for it in Study 1. In Study 2 each condition is compared with its own perspective baseline.
• Lure identity not stated. The Method does not say whether the Session 3 lure was the very photo shown after the movie. The Discussion implies it was ("an event they saw after their museum tour").
• Study 2 rests on a post hoc split. Its own main effect of reactivation condition was only marginal (p = .09). The replication rests on pooling the two studies and on a median split made after the fact.
• Trial counts are not reported. In Study 1, three quarters of 32 events is 24 reactivated and 8 baseline events before exclusions (our computation). The split between match and mismatch is not stated.
• Individual differences appear only through the reliving split. Whether a person's susceptibility is stable was not tested.
• Inconsistencies found:
  • Study 1's memory-type effect is reported as F(1, 39) = 25.16, p < .0001, ηp2 = .39, identical to Study 2's.
    • In Study 1, with 42 participants and all factors within-subject, the other effect of the same ANOVA has F(1, 41).
    • Probably a copy error.
  • Study 2 reports t(39) for confidence but t(40) elsewhere, with 41 participants.
  • Study 2's Session 2 accuracy (.85 against .84, SEM .03 each) is identical to Study 1's.
  • In Study 2, the novel photo in match trials is said to be "from the participant's own camera". But it shows a stop the participant did not visit, so it was presumably taken by another participant's camera.

What it means for Kurisutina

• Question 3: reactivation changes a person's memory in two directions at once.
  • Elements of the reminded event are recognised more (hits up).
  • Related material seen just after the reminder is later accepted as part of the event (false alarms up, by more than hits). In Study 2, confidence gained more for these false alarms than for hits (difference scores +0.52 against −0.01 on a 1–5 scale).
  • The better the reminder matches the experience, the larger both changes: correct order, own perspective, stronger reliving.
  • A replica whose store keeps source tags and does not change on retrieval would show neither effect (our inference).
  • One that retrieves and re-writes its memory (park2023's stream, with reflections) might show the false-alarm side, but no dependence on cue quality (a guess, untested).
  • Either way, reactivation is a place where person and replica are expected to part, and so a direct test of the candidate prediction.
• Reference measures for the planned test. Person and replica take in the same new material and are tested at delays with free recall, lures, misinformation and reactivation. The paradigm transfers directly.
  • Material. A camera-recorded experience gives the person's own cues and a record for scoring. The same photos (or their descriptions) are the replica's input.
  • Reactivation. At 48 h, cue half the events, some with matching cues (own order, own perspective) and some with mismatching ones. Follow each cue with one related novel item. Hold the other events back as baseline.
  • Test. At a further 48 h: recognition of targets and lures by condition, with confidence; for the person, also reliving.
  • Human reference. Table 1 and the effects above, from 83 young adults (our sum), four days after the tour.
  • Scoring, on pattern rather than accuracy:
    • a hit gain for reactivated events;
    • a larger false-alarm gain for post-reactivation items;
    • match above mismatch for both;
    • a confidence rise for false alarms but not hits;
    • the match effect growing with reliving.
  • Longer delays. This paper covers four days. hirst2009_911_memory supplies the human reference for months and years (consistency, confidence, transitions).
  • Misinformation. The novel photo here is related, true-to-the-place material presented as "is this related?", not as a claim about the event. A socially attributed misinformation phase (schacter2011, Edelson et al.) is a separate kind and needs its own reference (nichols2019).
• Elicitation: interviews reactivate memories and may change them (brief 6.2, 6.3, 7).
  • A few seconds of cueing sufficed when new material followed.
    • A cue of a few seconds (six photos at 1.25 s in Study 1), followed by related new material, changed what people recognised two days later.
    • Brief 6.3 notes that in the mouse brief cueing did not make memories labile. Whatever the mechanism, this human result means an interview that shows material after a cue cannot be assumed safe because the cue was short.
  • Log cues and their order (6.2). In Study 1 the order of the cues was itself the manipulation. In Study 2 the perspective was (own photos against others'). The 6.2 rule "log exactly which cues were presented, in what order" is supported. Also log whose photo or account each cue was.
  • Never show a model reconstruction and count later agreement (6.2). A model-generated reconstruction shown after cueing an episode is exactly the kind of related new item people later endorse as experienced, with a confidence gain larger than for true recognition (Study 2). Agreement after it is not independent confirmation.
  • Record reliving. A reliving rating per cue is a cheap index of how strongly the memory was reactivated, and people differ in it. Record it beside dwell time (6.2).
  • Hold back controls (7). The baseline events here are the "non-elicited control memories" of the perturbation audit. Holding back a random set of episodes from cueing gives the audit its counterfactual. It must measure both sides, hits and false alarms, for cued against uncued episodes, because reactivation also strengthens.
  • Loaded memories already carry interview effects. Memories collected over several interviews may already include material taken in during earlier sessions. The source field (5.3) should record what was shown before each answer.
• Individual differences are not settled. Only the post hoc reliving split speaks to them here. With nichols2019 (stable within a kind of distortion, not across), a person's susceptibility to reactivation-with-new-material has to be measured in this paradigm if the replica is to match it.

Cross-references

• summaries/carry_on/hirst2009_911_memory.md: how a real memory changes over months and years, with a retest check.
• summaries/carry_on/schacter2011_adaptive_distortion.md: by two of the same authors; Hupbach's reminder-plus-new- learning paradigm and the adaptive account of updating.
• summaries/carry_on/nichols2019_false_memory_tasks.md: susceptibility is stable within a paradigm, not across; confidence transfers more than errors do.
• summaries/carry_on/park2023_generative_agents.md: an LLM memory stream in which retrieval and reflection write back into memory.
• summaries/memory/johnson1993_source_monitoring.md (cited by the authors) and summaries/memory/lindsay1990_source_suggestions.md: source monitoring and post-event suggestion.
• summaries/memory/nader2000_reconsolidation.md, summaries/memory/schiller2010_reconsolidation.md, summaries/memory/chalkia2020_reconsolidation_replication.md: reconsolidation. The last is a registered replication that found no persistent effect in human fear memory; this paper's result does not depend on reconsolidation being the mechanism.
• summaries/memory/rudoy2009_targeted_reactivation.md: sound cues during a nap preceded better recall of the cued items (that summary cautions that better access is not proof of a strengthened trace).
• summaries/08_suzuki2004.md (brief [8]): re-exposure duration and lability, behind the dwell-time rule (6.2, 6.3).
• docs/research/elicitation_design_2026-09-22.md and docs/research/memory_retrieval_as_intervention_2026-09-22.md: earlier notes on elicitation and on retrieval as an intervention.