KurisutinaResearch date: 2026-09-22.
Citation and status. Dorothy Tse, Rosamund F. Langston, Masaki Kakeyama, Ingrid Bethus, Patrick A. Spooner, Emma R. Wood, Menno P. Witter, and Richard G. M. Morris. Schemas and memory consolidation. Peer-reviewed, Science 316(5821), 76–82; published 2007-04-06. DOI; PubMed.
Acquisition and reading. Read the entire main article, all figure legends, and references in the seven-page public academic copy. Read all 24 pages of Supporting Online Material, including methods, tables S1–S2, figures S1–S8, and references, in the 32-page combined public repository copy. Main figures and supplementary figures/tables were also inspected visually. Adjacent articles appearing on shared journal pages were excluded. No raw data, later replication, or subsequent debate paper was read for this note. The combined copy includes a publisher download cover dated 2011-08-11; both sources reproduce the 2007 article. No open redistribution license was identified. Local main PDF, main text, combined PDF, combined text, and provenance.
Source findings (factual paraphrase, approximately 150 words). Male rats gradually learned six stable flavor–place associations in an arena over weeks. Subsequent novel flavors paired with nearby new locations were learned during one training trial per association. Hippocampal lesions before initial training prevented normal acquisition. After established learning, lesions made 48 hours after new training spared later expression of new associations; lesions at three hours impaired it. Previously learned associations survived either delay. Further novel learning after lesions was impaired. In another experiment, new associations were remembered after stable flavor–place training but not after a history in which the mappings repeatedly changed, despite familiarity with the same task components. The measured outcome was allocation of digging during cued probes, not verbal report or a direct assay of a cortical trace. The authors interpret these results as unusually rapid systems consolidation enabled by a schema. The experiments establish training-history and lesion-timing effects on behavior; they do not directly identify the storage site or the representational format transferred between brain systems.
Methodological audit — additional source facts that change interpretation.
- “One trial” contained three rewarded collection trips. Probe scoring used an unrewarded 120-second window, followed by three half-pellet rewards to limit extinction. Experiment 2's first postoperative novel-association probe tested the counterpart not probed before surgery; later pooled endpoints included both testing histories. Experiment 3 probed a novel association before surgery at both delays. Its postoperative behavioral tests followed 14 days of recovery. These schedules matter when defining a single-exposure learning claim.
- Acquisition used a search-error performance index with chance 50; probe digging proportion had chance 1/6. The former is not percentage first-choice accuracy. Methods specify scheduled training, not a mastery threshold or performance-based selection. Experiment 3 reports all rats above chance at the presurgical novel probe, without showing those data.
- Lesions were extensive but incomplete: group mean spared hippocampal tissue ranged from about 10% to 19% (table S2). No cortical recording or cortical inactivation localized the retained information. Permanent lesions also perturb interacting networks; behavioral survival does not by itself demonstrate a new cortical trace formed between three and 48 hours.
- Experiment 3 has unresolved reporting inconsistencies: main text says 18 rats; supplement says 20. Figure 4 assigns seven lesioned rats to three hours and six to 48 hours; supplementary methods and figure S8 give six and seven respectively. The supplement explains that some initial controls later underwent a second learning/surgery phase: 18 control records are not 18 independent control animals. Resolve these details before quantitative reanalysis.
- In Experiment 4, the unstable condition remapped flavor–place pairings every two sessions. Novel associations were arbitrary bindings within the familiar task and spatial framework. The experiment did not compare rule-congruent additions with explicit exceptions or test content from a different domain. Room assignment was counterbalanced; training-history differences and possible effects on attention or expected learnability remain relevant to interpretation.
Interpretation for extraction and transfer — our reasoning. A memory record needs an interpreter with appropriate prior knowledge. Here, fast learning followed extensive acquisition of a cue vocabulary, spatial framework, reward structure, and reliable task relations. A recipient supplied only the latest event cannot be assumed to use it as the trained source would. Conversely, prior knowledge cannot determine a genuinely new arbitrary binding: the particular flavor–location observation still carries indispensable information.
The result motivates testing interactions between recipient knowledge and transferred evidence. It does not establish that a biological schema is an explicit graph, that all familiar-domain content becomes rapidly independent of the hippocampus, or that stable prior knowledge specifically changes consolidation rather than encoding or access. An archive, a learned rule system, and a context-sensitive hybrid could all produce some of the reported endpoints. Their predictions must be compared under new updates and exceptions.
Falsifiable functional-transfer experiment — proposal. In a controlled invented domain, randomize source participants to stable structured relations, stable arbitrary relations, or volatile mappings, while matching exposure and feedback. Record initial skill and uncertainty instead of selecting only successful learners. Add three distinct item classes: genuinely new arbitrary bindings, predictable consequences of existing relations, and explicit exceptions. Assign separate matched families to immediate and delayed tests, with a delayed-only control, because probes themselves supply practice and potentially new learning.
Give recipient systems the same observed training evidence and new-event records. Compare an episode archive, rules alone, and a hybrid with matched update opportunities. Cross recipient prior framework with transferred content: matching knowledge, a conflicting framework, and a novice state. Score the recipient's own choices, errors, retention, interference with old bindings, and subsequent learning across context changes. Source-person forecasting is a separate endpoint; prediction accuracy alone is not successful behavioral transfer.
The central prediction is an interaction: matching prior knowledge should improve use of a small new record without replacing the need for arbitrary new content. A rules-only recipient that handles predictable cases but misses exceptions fails the binding test; an archive that retrieves facts but fails to update future choices fails the functional-use test. If a common update rule plus personal prior state matches a personalized rule, the experiment supports transfer of learned state rather than identification of a unique personal learning mechanism. Delayed performance differences alone would still not establish biological systems consolidation.