Search

Research on working memory (WM) has followed two largely independent traditions: One concerned with memory for sequentially presented lists of discrete items, and the other with short-term maintenance of simultaneously presented arrays of objects with simple, continuously varying features. Here we present a formal model of WM, the interference model (IM), that explains benchmark findings from both traditions: The shape of the error distribution from continuous reproduction of visual features, and how it is affected by memory set size; the effects of serial position for sequentially presented items, the effect of output position, and the intrusion of nontargets as a function of their distance from the target in space and in time. We apply the model to two experiments combining features of popular paradigms from both traditions: Lists of colors (Experiment 1) or of nonwords (Experiment 2) are presented sequentially and tested through selection of the target from a set of candidates, ordered by their similarity. The core assumptions of the IM are: Contents are encoded into WM through temporary bindings to contexts that serve as retrieval cues to access the contents. Bindings have limited precision on the context and the content dimension. A subset of the memory set--usually one item and its context--is maintained in a focus of attention with high precision. Successive events in an episode are encoded with decreasing strength, generating a primacy gradient. With each encoded event, automatic updating of WM reduces the strength of preceding memories, creating a recency gradient and output interference.

Drift-Diffusion Models (DDMs) are a widely-used class of models that assume an accumulation of evidence during a quick decision. These models are often used as measurement models to assess individual differences in cognitive processes, such as an individual's evidence accumulation rate and response caution. An additional underlying assumption of these models is that there is internal noise in the evidence accumulation process. However, fitting DDMs to experimental choice-response time data alone cannot yield estimates of an individual's evidence accumulation rate, caution, and internal noise at the same time. This is due to an intrinsic joint-unidentifiability of these parameters when fitting DDMs to behavioral data. We introduce methods of estimating these parameters at the same time with additional data types. The methods to estimate model parameters rely on Bayesian inference and simulation-based Bayesian inference. We show why these methods are useful without making strong assumptions.

One of the best-known demonstrations of long-term learning through repetition is the Hebb effect: Immediate recall of a memory list repeated amidst nonrepeated lists improves steadily with repetitions. However, previous studies often failed to observe this effect for visuospatial arrays. Souza and Oberauer (2022) showed that the strongest determinant for producing learning was the difficulty of the test: Learning was consistently observed when participants recalled all items of a visuospatial array (difficult test) but not if only one item was recalled, or recognition procedures were used (less difficult tests). This suggests that long-term learning was promoted by increased testing demands over the short term. Alternatively, it is possible that lower testing demands still lead to learning but prevented the application of what was learned. In four preregistered experiments (N = 981), we ruled out this alternative explanation: Changing the type of memory test midway through the experiment from less demanding (i.e., single item recall or recognition) to a more demanding test (i.e., full item recall) did not reveal hidden learning, and changing it from the more demanding to a less demanding test did not conceal learning. Mixing high and low demanding tests for nonrepeated arrays, however, eventually produced Hebb learning even for the less demanding testing conditions. We propose that testing affects long-term learning in two ways: Expectations of the test difficulty influence how information is encoded into memory, and retrieval consolidates this information in memory.

The Hebb repetition effect shows improvement in serial recall of repeated lists compared to random nonrepeated lists. Previous research using simple span tasks found that the Hebb repetition effect is limited to constant uninterrupted lists, suggesting chunking as the mechanism of list learning. However, the Hebb repetition effect has been found in complex span tasks, which challenges the chunking explanation, as successive list items are separated by distractor processing, possibly interfering with the unified representations. We tested the possibility that Hebb repetition learning arises from chunking in simple span, but from position--item associations in complex span. In a series of five experiments, we found evidence that contradicts that hypothesis. Results show that (a) Hebb repetition learning in a complex span task can be transferred to a simple span task; (b) Hebb repetition learning from a complex span task cannot be transferred to a partially repeated simple span task; (c) partial repetition in a complex span task does not lead to learning; (d) Hebb repetition learning from a simple span task can be transferred to a complex span task; and (e) repeating the distractors in complex span has no impact on the Hebb repetition effect. These results suggest that the mechanism underlying the Hebb repetition effect in simple and complex span tasks is the same and points at the creation of chunks while excluding the distractors from the long-term memory representation.

Recall performance in working memory (WM) is strongly affected by the similarity between items. When asked to encode and recall list of items in their serial order, people confuse more often the position of similar compared to dissimilar items. Models of WM explain this deleterious effect of similarity through a problem of discriminability between WM representations. In contrast, when lists of items that are all semantically similar are compared to lists of dissimilar items, semantic similarity does not negatively impact order memory, questioning the idea that semantic information is part of the WM content. This study reports four experiments in which semantic similarity was manipulated using lists composed of multiple semantic categories. These experiments revealed two main patterns. First, semantic similarity can "increase", rather than decrease, order memory. Second, semantic knowledge reliably constrains the way items migrate; when migrating, items tend to do so more often toward the position of other similar items, rather than migrating toward other dissimilar items. These results challenge the way current models of WM represent similarity. The plausibility of different theoretical accounts and mechanisms is discussed.

This work investigates how people make judgments about the content of their visual working memory (VWM). Some studies on long-term memory suggest that people base those metacognitive judgments on the outcome of a retrieval attempt. In contrast, Son and Metcalfe (2005) observed that people identify poorly remembered items immediately, presumably by the lack of familiarity for the retrieval cue. We tested these two hypotheses in the context of metacognition in VWM. In three experiments, we investigated participants' response behavior in a color reproduction task with a hidden color wheel. With this procedure, participants must search for the intended response, starting from a random color. We assumed that instant awareness of the inability to retrieve an information would be reflected in selecting the first, random color, rather than search for a particular color in the wheel. Although participants provided a substantial number of low-confidence responses, results of an adapted mixture modeling analysis yielded little evidence for quick guesses. Rather, participants consistently searched for a color (even with unfamiliar retrieval cues in Experiment 2), and only quickly guessed when being cued with objects at test that were not previously presented (Experiment 3). We conclude that people usually engage in retrieval attempts for providing judgments about their VWM, even when information is poorly remembered.

Longer free time between presentation of verbal list items often leads to better immediate serial recall. The present series of three experiments demonstrates that this beneficial effect of time is more general than has been known: It is found for verbal items presented visually and auditorily (Experiments 1 and 2), and also when people engage in concurrent articulation during presentation, thereby preventing rehearsal (Experiment 3). The effect of time is to improve memory most strongly for the later part of the list, contrary to what is predicted from the assumption that time between items is used to bolster memory traces of already encoded items through rehearsal, refreshing, or elaboration. The data are compatible with a ballistic form of short-term consolidation, and with the assumption that encoding an item into working memory partially depletes a limited resource, which is replenished over time.

Both the experimental and the psychometric investigation of the WM capacity limit depend critically on the assumption that performance in our tests of WM reflects that capacity limit to a good approximation. Most tasks to measure WM rely on testing memory after a short time during which participants are asked to maintain information in WM. In these tests, episodic long-term memory is likely to also lay down a trace of the memory set. Therefore, participants can draw on two sources of information when memory is tested, making it difficult to separate the contributions of WM and episodic LTM to the performance on immediate-memory tests. Here we use proactive interference to distinguish between these two sources of remembered information, building on the fact that episodic memory is vulnerable to proactive interference, whereas WM is protected against it. We use a release-from-PI paradigm to determine the extent to which commonly used WM tasks reflect contributions from episodic LTM. We focus on memory for serial order of verbal lists, but also include visual and spatial WM tasks. The results of five experiments demonstrate that although some tasks used to investigate WM are heavily contaminated by episodic LTM, other popular paradigms such as serial and probed recall, and the standard version of the continuous color-reproduction task, are not. Measuring proactive interference can help researchers determine the extent to which WM and episodic LTM contribute to performance in immediate-memory tasks.

Research on working memory (WM) has followed two largely independent traditions: One concerned with memory for sequentially presented lists of discrete items, and the other with short-term maintenance of simultaneously presented arrays of objects with simple, continuously varying features. Here we present a formal model of WM, the interference model (IM), that explains benchmark findings from both traditions: The shape of the error distribution from continuous reproduction of visual features, and how it is affected by memory set size; the effects of serial position for sequentially presented items, the effect of output position, and the intrusion of nontargets as a function of their distance from the target in space and in time. We apply the model to two experiments combining features of popular paradigms from both traditions: Lists of colors (Experiment 1) or of nonwords (Experiment 2) are presented sequentially and tested through selection of the target from a set of candidates, ordered by their similarity. The core assumptions of the IM are: Contents are encoded into WM through temporary bindings to contexts that serve as retrieval cues to access the contents. Bindings have limited precision on the context and the content dimension. A subset of the memory set—usually one item and its context—is maintained in a focus of attention with high precision. Successive events in an episode are encoded with decreasing strength, generating a primacy gradient. With each encoded event, automatic updating of WM reduces the strength of preceding memories, creating a recency gradient and output interference. [ABSTRACT FROM AUTHOR]

Novelty-gated encoding is the assumption that events are encoded more strongly into memory when they are more novel in comparison to previously encoded events. It is a core assumption of the SOB model of serial recall (Farrell & Lewandowsky, 2002). We present three experiments testing some predictions from novelty-gated encoding. Experiment 1 shows that the probability of recalling the third item in a list correctly does not depend on whether it is preceded by phonologically similar or dissimilar items. Experiment 2 shows that in lists of items from three classes (nonwords, spatial locations, and abstract drawings) the probability of recalling an item does not depend on whether it is preceded by items from the same or another class. Experiment 3 used a complex-span paradigm varying the phonological similarity of words that are read aloud as distractors in between memory items. Contrary to a prediction from novelty-gated encoding, similar distractors did not impair memory more than dissimilar distractors. The results question the assumption of novelty-gated encoding in serial recall. We discuss alternative explanations for the phenomena that this assumption has previously helped to explain. The present evidence against novelty-gated encoding might point to boundary conditions for the role of prediction error in the acquisition of memories.