In this work, we propose a neural network model for free recall that draws direct parallels between neural machine translation (NMT) and cognitive models of memory search, specifically the Context Maintenance and Retrieval (CMR) model. We hypothesize that NMT advancements such as attention mechanisms (Luong et al., 2015) closely resemble how humans reactivate prior contexts (“mental time travel”; Tulving, 1985). To demonstrate these parallels, we train a seq2seq model with attention as a cognitive model of memory search and evaluate behavior against human free recall data. We find that the model can capture typical free recall patterns previously observed (Kahana et al., 2022); and after optimization, the model demonstrates the same optimal behavior as previously derived by the CMR model (Zhang, Griffiths, & Norman, 2023). Performing an ablation study, we demonstrate that behavioral differences between models with and without attention align with impaired behavior observed in hippocampal amnesia patients.

We compare storytelling in GPT-3.5, a recent large language model, with human storytelling. We hypothesized that GPT differs from humans in the kind of memories it possesses, and thus could perform differently on tasks influenced by memory, such as storytelling. We used an existing dataset of human stories, either recalled or imagined (Sap et al., 2022), and generated GPT stories with prompts designed to align with human instructions. We found that GPT's stories followed a common narrative flow of the story prompt (analogous to semantic memory in humans) more than details occurring in the specific context of the event (analogous to episodic memory in humans). Furthermore, despite lacking episodic details, GPT-generated stories exhibited language with greater word affect (valence, arousal, and dominance). When provided with examples of human stories (through few-shot prompting), GPT was unable to match its stories' narrative flow or affective aspects with human stories.

How people represent categories—and how those representations change over time—is a basic question about human cognition. Previous research has suggested that people categorize objects by comparing them to category prototypes in early stages of learning but use strategies that consider the individual exemplars within each category in later stages. However, many category learning experiments do not accurately reflect the environmental statistics of the real world, where the probability that we encounter an object changes over time. Our goal in this study was to introduce memory constraints by presenting each stimulus at intervals corresponding to the power-law function of memory decay. Since the exemplar model relies on the individual’s ability to store and retrieve previously seen exemplars, we hypothesized that adding memory constraints that better reflect real environments would favor the exemplar model more early on compared with later. Confirming our hypothesis, the results illustrate that under realistic environmental statistics with memory constraints, the exemplar model’s advantage over the the prototype model decreases over time.

The method of loci is a powerful mnemonic technique for memorizing a list of unrelated items. With a pre-specified routein a familiar ”memory palace”, one can encode material by attaching items to loci along this route, and later effectivelyrecall them by mentally walking along the same route. Despite its efficacy, there is no existing model that explains whythe method of loci promotes memory improvement during memory search. To fill this gap, we provide a rational accountof why the method of loci improves memory. We define memory search as a task with the goal of minimizing retrievalcost, and demonstrate that the method of loci gives an optimal policy for this task. We discuss the implications of thisresult, and compare it with models of memory search without using mnemonic techniques.

Contrary to common intuition, groups of people recalling information together remember less than the same number of individuals recalling alone (i.e., the collaborative inhibition effect). To understand this effect in a free recall task, we build a computational model of collaborative recall in groups, extended from the Context Maintenance and Retrieval (CMR) model which captures how individuals recall information alone (Polyn, Norman, & Kahana, 2009). We propose that in collaborative recall, one not only uses their previous recall as an internal retrieval cue, but also listens to someone else's recall and uses it as an external retrieval cue. Attending to this cue updates the listener's context to be more similar to the context of someone else's recall. Over an existing dataset (Gates, Suchow, & Griffiths, 2022), we show that our model successfully captures the collaborative inhibition effects, as well as additional recall patterns such as recency and semantic clustering effects.

Hindsight bias is exhibited when knowledge of an outcome (i.e., an anchor) affects subsequent recollections of previous predictions (i.e., an estimate). Hindsight bias usually leads to estimates being remembered as closer to the anchor than they actually were. The exact amount of hindsight bias exhibited depends on the anchor value and the anchor plausibility, with experimental results showing that hindsight bias is elicited only when the anchor is perceived to be plausible. In this paper we present a Bayesian model that captures the relationship between hindsight bias and anchor plausibility. This model provides a rational account of hindsight bias by considering memory recall as a statistical problem, where the goal is to reconstruct the original estimate using the anchor as new evidence. Simulations show that the modeled trends align closely with previously published human data.

We provide new evidence concerning two views of episodic associations: The independent associations hypothesis (IAH)posits that associations are unidirectional and separately modifiable links (AB; AB); the associative symmetry hypothesis(ASH) considers the association to be a holistic conjunction of A and B representations. While existing literature focuseson tests that compare the correlation of forward and backward associations and favors ASH over IAH, we provide thefirst direct evidence of IAH by showing that forward and backward associations are separately modifiable for semanti-cally related pairs. In two experiments, participants studied 30 semantically unrelated and 30 semantically related pairsintermixed in a single list, and then performed a series of up to eight cued-recall test cycles. All pairs were tested in eachcycle, and the testing direction (A-? or B-?) alternated between cycles. Consistent with prior research, unrelated pairsexhibited associative symmetry accuracy and response times improved gradually on each test, suggesting that testing inboth directions strengthened the same association. In contrast, semantically related pairs exhibited a stair-like pattern,where performance did not change from odd to even tests when the test direction changed; it only improved between testsof the same direction. We conclude that episodic associations can have either a holistic representation (ASH) or separatedirectional representations (IAH), depending on the semantic relatedness of their constituent items.

In this study, we investigated the neural correlates of a visual working memory task. Two experiments were carriedout using scalp electroencephalography (EEG) and Electrocorticography (ECoG), respectively. In each trial, participants judgedwhether a test face had been among a small set of recently studied faces. We used a combination of hidden semi-Markov models(HSMMs) and multi-variate pattern analysis (MVPA) to decompose the neural signal into a sequence of latent stages. Analyzedseparately, EEG and ECoG data yielded converging results on the durations of recovered stages. Combining these stages withthe high spatial resolution of ECoG suggested that activity in the temporal cortex reflected item familiarity in the retrievalstage; and that once retrieval is complete, there is active maintenance of the studied face set in the medial temporal lobe (MTL).During this same period, the frontal lobe guides the decision by means of theta coupling with the MTL.

Recent brain imaging studies have provided new insight into how students are able to extend their previous problem solving skills to new but similar problems. It is still unclear, however, what the basis is of individual differences in their success at transfer. In this study, 75 subjects had been trained to solve a set of mathematical problems before they were put into the fMRI scanner, where they were challenged to solve modified versions of familiar problems. A hidden semi-Markov model identified the sequential structure of thought when solving the problems. Analyzing the patterns of brain activity over the sequence of states identified by the model, we observed that subjects who showed consistent brain patterns performed better. This consistency refers to both how consistently subjects respond to different problems (within-subject consistency), and how brain responses of a given subject deviate from the population average (between-subjects consistency). Early withinsubject consistency is particularly predictive of later performance in the experiment.

Executive function (EF), its importance for scholastic achievement and the question of whether or not EF is malleable, have become a topic of intense interest. Education or schooling is often seen as effective approaches to enhance EF due to the specific school-related requirements as compared to kindergarten or pre-school. However, no study to date has investigated whether targeted training focusing on those domains might be comparable with regular schooling in improving EF and fluid intelligence (Gf). The aim of the present study was to replicate and extend the previously demonstrated schooling effects on EF by using a school-cutoff design, and to further investigate whether a theoretically motivated intervention targeting specific EF, i.e., working memory (WM) or inhibitory control (IC), could achieve comparable effects with schooling in both, WM and IC, as well as Gf. 91 6-year-old kindergarteners and first-graders with similar chronological age participated the study. We compared the performance of a first-grade schooling group with that of two kindergarten training groups as well as a business-as-usual kindergarten control group. Participants were assessed in WM, IC and Gf at baseline, immediately after the intervention (posttest), as well as 3 months after training completion (follow-up). The results showed that the schooling group indeed outperformed the kindergarten groups at baseline in several cognitive tasks. Furthermore, both the WM and IC training showed pronounced gains in the trained tasks, as well as varying degrees of improvement in non-trained outcome measures. Most importantly, both training groups achieved comparable performance with the schooling group, which was especially apparent in Gf at follow-up. Our findings provide further evidence for the malleability of EF demonstrating that both, long-term and short-term interventions can facilitate the acquisition of those important skills, and as such, our work has important implications for educational practice.