Memory Mechanisms

When better-studied items ("strong") are mixed with worse-studied items ("weak"), episodic recognition judgements are expected to produce a greater difference due to strength than when lists contain items of just one strength (pure lists). Such list-strength effects are found for reading aloud but not spaced repetition. An inverted effect is found for stimulus duration when presentation times are short (~ 500 ms, reinforced by our recent findings; Caplan and Guitard, 2024, 2025, in press). To explain near-null list-strength effects, the Bayesian, local-trace, Retrieving Effectively from Memory model (REM; Shiffrin & Steyvers, 1997) incorporates differentiation, whereby a strong item provides more evidence it was a target, but provides more diagnostic evidence against a lure. Differentiation produces a force toward inverted list-strength effects, offsetting an underlying upright effect. We found simulations using the original parametrization of REM robustly produces near-null effects, even when the model encodes additional traces upon repetitions. Further explorations of the parameter space (feature distribution, feature copy error rate and vector dimensionality) show that the model has a number of ways to produce inverted list-strength effects, a previously unsung strength of REM, but also sizeable upright effects. Grid-search fits to our data, however, proved challenging. REM comes close to hit and false-alarm rate values but misses rank-orders of the measures in both datasets. Inspection of model output suggests there is a strong constraint in REM to produce large mirror effects, yoking hit and false-alarm rates together, whereas in these fast-presentation-rate data, these are relatively uncoupled. We discuss future modifications that might make REM a little less effective while largely retaining its spirit.

This is an in-person presentation on July 18, 2026 (09:00 ~ 09:20 EDT).

Everyday memory experiences often involve continuous interactions with environmental cues. Yet our current understanding of human memory is predominantly based on controlled laboratory experiments where participants engage in memory tasks in isolation. In this work, we build a computational model to understand how external cues help people continue their recall, accounting for how people search their memories within and across events. Our model simulations demonstrate that beneficial cues target events with the most remaining information, as cues tap into previously unexplored memory spaces. Furthermore, information at event boundaries serves as a superior cue, aligning with prior empirical and computational work suggesting that boundaries act as entry points into past experiences. We test our model predictions over two free recall experiments, ranging from simplified word lists to more naturalistic continuous narratives.

This is an in-person presentation on July 18, 2026 (09:20 ~ 09:40 EDT).

Episodic memory can recall detailed event representations from sparse cues, yet information learned during an event can also be generalized to novel situations. To explain this flexibility, many accounts posit multiple memory systems or the storage of separate traces for specific and generalized content. Here, we argue that single-system, single-trace architectures may be substantially more flexible than typically assumed. We show that global matching models of episodic memory and exemplar models of classification can be viewed as static approximations of a recurrent network architecture that we call ATHENA. Simulations demonstrate that plausible modulations of network components alter the effective level of competition between traces during retrieval. In traditional cognitive models, this corresponds to dynamically varying the width of the similarity kernel. Allowing this variation enables a single-system, single-trace model to retrieve individual event memories (i.e., exemplars), averages of related memories (i.e., prototypes or clusters), and global averages across all memories (i.e., base rates of features) using the same memory traces and retrieval cue. We conclude that, once static-time assumptions are relaxed, episodic memory systems can naturally support both precise recall and flexible generalization.

This is an in-person presentation on July 18, 2026 (09:40 ~ 10:00 EDT).

A major debate among memory theorists is whether there is a threshold process in memory, such that memory can fail entirely. Results are conflicting, which has been attributed to the fact that memory responses have been confounded by decision processes in prior work. To more directly index underlying memory uncontaminated by decision processes, we examined source-memory-driven eye movements during visual search for targets in real-world scenes. Across two studies (n=84 and n=68), we found that memory-driven eye movement accuracy demonstrated the pattern predicted by threshold theories: it was well fit by a mixture model consisting of precision and guessing distributions, such that memory was either highly precise or failed entirely. This threshold mixture model provided a better fit than the opposing non-threshold model consisting of a single continuous distribution. These results also held when only confidently recognized scenes were analyzed, which had a false alarm rate of <1% and were thus highly diagnostic of successful item recognition. Therefore, the results could not be explained by encoding failures or by item recognition failures inducing artifactual guessing. Together, this evidence for a threshold process in memory supports threshold theories (e.g., DPSD) and conflicts with continuous process theories (e.g., SDT) of recognition memory.

This is an in-person presentation on July 18, 2026 (10:00 ~ 10:20 EDT).