new research on working memory

Scientists Pinpoint the Uncertainty of Our Working Memory

The human brain regions responsible for working memory content are also used to gauge the quality, or uncertainty, of memories, a team of scientists has found, uncovering how these neural responses allow us to act and make decisions based on how sure we are about our memories.

New Study Shows the Extent We Trust Our Memory in Decision-Making

The human brain regions responsible for working memory content are also used to gauge the quality, or uncertainty, of memories, a team of scientists has found. Its study uncovers how these neural responses allow us to act and make decisions based on how sure we are about our memories.

“Access to the uncertainty in our working memory enables us to determine how much to ‘trust’ our memory in making decisions,” explains Hsin-Hung Li, a postdoctoral fellow in New York University’s Department of Psychology and Center for Neural Science and the lead author of the paper , which appears in the journal Neuron . “Our research is the first to reveal that the neural populations that encode the content of working memory also represent the uncertainty of memory.”

Working memory, which enables us to maintain information in our minds, is an essential cognitive system that is involved in almost every aspect of human behavior—notably decision-making and learning. 

For example, when reading, working memory allows us to store the content we just read a few seconds ago while our eyes keep scanning through the new sentences. Similarly, when shopping online, we may compare, “in our mind,” the item in front of us on the screen with previous items already viewed and still remembered. 

“It is not only crucial for the brain to remember things, but also to weigh how good the memory is: How certain are we that a specific memory is accurate?” explains Li. “If we feel that our memory for the previously viewed online item is poor, or uncertain, we would scroll back and check that item again in order to ensure an accurate comparison.”

While studies on human behaviors have shown that people are able to evaluate the quality of their memory, less clear is how the brain achieves this. 

More specifically, it had previously been unknown whether the brain regions that hold the memorized item also register the quality of that memory.

In uncovering this, the researchers conducted a pair of experiments to better understand how the brain stores working memory information and how, simultaneously, the brain represents the uncertainty—or, how good the memory is—of remembered items. 

In the first experiment, human participants performed a spatial visual working memory task while a functional magnetic resonance imaging (fMRI) scanner recorded their brain activity. For each task, or trial, the participant had to remember the location of a target—a white dot shown briefly on a computer screen—presented at a random location on the screen and later report the remembered location through eye movement by looking in the direction of the remembered target location.

Here, fMRI signals allowed the researchers to decode the location of the memory target—what the subjects were asked to remember—in each trial. By analyzing brain signals corresponding to the time during which participants held their memory, they could determine the location of the target the subjects were asked to memorize. In addition, through this method, the scientists could accurately predict memory errors made by the participants; by decoding their brain signals, the team could determine what the subjects were remembering and therefore spot errors in their recollections.  

In the second experiment, the participants reported not only the remembered location, but also how uncertain they felt about their memory in each trial. The resulting fMRI signals recorded from the same brain regions allowed the scientists to decode the uncertainty reported by the participants about their memory. 

Taken together, the results yielded the first evidence that the human brain registers both the content and the uncertainty of working memory in the same cortical regions.

“The knowledge of uncertainty of memory also guides people to seek more information when we are unsure of our own memory,” Li says in noting the utility of the findings.

The study’s other researchers included Wei Ji Ma and Clayton Curtis, professors in NYU’s Department of Psychology; Thomas Sprague, an NYU postdoctoral researcher at the time of the study and now an assistant professor at the University of California, Santa Barbara; and Aspen Yoo, an NYU doctoral student at the time of the study and now a postdoctoral fellow at the University of California, Berkeley.

The research was supported by grants from the National Eye Institute (NEI) (R01 EY-016407, R01 EY-027925, F32 EY-028438) and the NEI Visual Neuroscience Training Program (T32-EY007136).

Press Contact

Attention and working memory: Two sides of the same neural coin?

In 1890, psychologist William James described attention as the spotlight we shine not only on the world around us, but also on the contents of our minds. Most cognitive scientists since then have drawn a sharp distinction between what James termed “sensorial attention” and “intellectual attention,” now usually called “attention” and “working memory,” but James saw them as two varieties of the same mental process. 

New research by Princeton neuroscientists suggests that James was on to something, finding that attention to the outside world and attention to our own thoughts are actually two sides of the same neural coin. What's more, they have observed the coin as it flips inside the brain. 

A paper published in Nature on March 31 by Matthew Panichello , a postdoctoral research associate at the Princeton Neuroscience Institute, and Timothy Buschman , an assistant professor of psychology and neuroscience at Princeton, found that attention and working memory share the same neural mechanisms. Importantly, their work also reveals how neural representations of memories are transformed as they direct behavior.

“When we act on sensory inputs we call it ‘attention,’” said Buschman. “But there’s a similar mechanism that can act on the thoughts we hold in mind.”

Timothy Buschman

In a pair of experiments with two rhesus macaque monkeys, the researchers found that neurons in the prefrontal cortices that focus attention on sensory stimuli are the very same ones that focus on an item in working memory. What's more, Panichello and Buschman actually observed the neural representations of those memories realigning in the brain as the monkeys selected which memories to act upon.

In one experiment, each monkey was seated before a computer monitor and a camera that tracked their eye movements. The monitor displayed pairs of randomly selected colored squares, one above the other. Then the squares vanished, requiring the monkey to remember the color and location of the squares. After a brief pause, a symbol appeared, telling the monkey which square they should select from their working memory. Then, after another pause, they reported the color of the selected square by matching it to a color wheel. 

To perform the task, each monkey needed to hold both colors in their working memory, select the target color from memory, and then report that color on the color wheel. After each response, the monkey was rewarded with droplets of juice. The closer their report was to the target color, the more droplets they earned. 

In a second experiment, to compare the selection of items from working memory to a more classic attention task, the researchers indicated the direction to the monkeys before they saw the colored squares. This allowed the macaques to focus all their attention on the indicated square (and ignore the other one). As expected, the monkeys performed better on this task because they knew in advance which square to attend to and which to ignore.

The researchers recorded neural activity in the prefrontal cortex, parietal cortex and visual cortex. The prefrontal cortex is associated with a variety of executive function processes including attention, working memory, planning and inhibition. In this study, the researchers discovered that the same neurons in the prefrontal cortex that directed attention were also used to select an item from the monkey's working memory.

Princeton neuroscientists have discovered that attention and working memory are much more closely connected than most modern cognitive scientists realized. They performed two experiments in which monkeys were shown two color blocks and a symbol that directed them to look at the top one (a circle or an upward slanted line) or the bottom one (a triangle or a downward slanted line). They then matched the selected color to its spot on the color wheel. In the first experiment (left), they saw the blocks first and then the directional signal. In the second (right), they saw the directional signal first and then the color blocks.

This wasn’t true everywhere in the brain. In an area in the visual cortex associated with color recognition and in an area in the parietal lobe associated with visual and spatial analysis, the processes of attending to sensory input and selecting the target color from working memory involved distinct neural mechanisms. 

“Attention allows you to focus your resources on a particular stimulus, while a similar selection process happens with items in working memory,” said Buschman. “Our results show the prefrontal cortex uses one representation to control both attention and working memory.”

The same neural recordings also showed how selecting an item changes memories so that they are either hidden away in working memory or used to make a response. This involves dynamically rotating the memory representation in the prefrontal cortex. 

This can be likened to holding a piece of paper with text on it. If you hold the paper edge-on to your face, you can’t read it. This concealment, Buschman explained, prevents the brain from triggering the wrong response, or triggering a response too early.

“The brain is holding information in a way that the network can’t see it,” he said. Then, when it came time to respond at the end of the trial, the memory representation rotated. Just as rotating the paper allows you to read and act upon the text, rotating the neural representation allows the brain to direct behavior.

“This dynamic transformation just blew me away,” said Buschman. “It shows how the brain can manipulate items in working memory to guide your action.”

Data from the first experiment: The spectrum of possible colors for the two blocks (upper and lower) are represented as a ring in the activity of neurons in the prefrontal cortex. When the animal is remembering both items (before selecting the target), these rings lie on separate “planes” within the brain. These planes are perpendicular to one another to keep the items separate. When one of the items is selected, the color rings rotate in order to align the colors for either item. This allows the brain to “read out” the color of the selected item, regardless of whether it was originally the upper or lower item.

“It is an important paper,” said Massachusetts Institute of Technology neuroscientist Earl Miller, who was not involved in this research. “Attention and working memory have often been discussed as being two sides of the same coin, but that has mainly been lip service. This paper shows how true this is and also shows us the ‘coin’ — the coding and control mechanisms that they share.”

“Our goal is not to overwrite the word ‘attention,’” said Buschman. Instead, he hopes that findings from decades of research on attention can be generalized to shed light on other forms of executive function. “Attention has been well-studied as the cognitive control of sensory inputs. Our results begin to broaden these concepts to other behaviors.”

“ Shared mechanisms underlie the control of working memory and attention ,” by Matthew F. Panichello and Timothy J. Buschman, appears in the March 31 issue of the journal Nature ( DOI: 10.1038/s41586-021-03390-w ) . This research was supported by the National Institute for Mental Health (R01MH115042 to TJB) and the Department of Defense ( a National Defense Science and Engineering Graduate Fellowship to MFP).

Related Stories

Princeton study reveals the brain's mysterious switchboard operator ..

Princeton University researchers report that a mysterious region deep in the human brain could be where we sort through the onslaught of stimuli from the outside world and focus on the information most important to our behavior and survival.

Princeton scientists identify neural activity sequences that help form memory, decision-making .

Princeton University researchers have used a novel virtual reality and brain imaging system to detect a form of neural activity underlying how the brain forms short-term memories that are used in making decisions.

Study shows how the brain pays attention .

The best-laid plans: how we update our goals based on new information ..

Humans are adept at setting goals and updating them as new situations arise — for example, a person who is playing a video game may switch to a new goal when their phone rings. Now, Princeton University researchers have identified mechanisms that govern how the brain incorporates information about new situations into our existing goals, according to research recently published in the Proceedings of the National Academy of Sciences (PNAS).

Hasson brings real life into the lab to examine cognitive processing .

Princeton University neuroscientist Uri Hasson strives to make research conditions in his lab as true to real life as possible, using uncommon subject matter —including slapstick comedy and high-school melodrama — in his studies. Hasson, an assistant professor in the Department of Psychology and the Princeton Neuroscience Institute, is exploring the underlying neural mechanisms of both the processes that allow the brain to integrate information over time and those that facilitate communication between people. 

Bad decisions arise from faulty information, not faulty brain circuits .

Princeton University researchers have found that bad decisions might be the fault of faulty information, rather than errors in the brain's decision-making process.

Out of the mouths of primates, facial mechanics of human speech may have evolved .

Two recent studies based at Princeton University suggest that the oral-facial component of human speech evolved from lip smacking, a friendly back-and-forth gesture performed by primates such as chimpanzees, baboons and macaques. The studies suggest a separate neural control for facial mechanics in primates that could help illuminate the neurological basis of speech disorders in humans.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings
• My Bibliography
• Collections
• Citation manager

Save citation to file

Email citation, add to collections.

• Create a new collection
• Add to an existing collection

Add to My Bibliography

Your saved search, create a file for external citation management software, your rss feed.

• Search in PubMed
• Search in NLM Catalog
• Add to Search

EXPRESS: Towards theoretically understanding how long-term memory semantics can support working memory performance

Affiliations.

• 1 Department of Psychological Sciences & Health, University of Strathclyde, 40 George Street, Glasgow, G1 1QE, UK.
• 2 Department of Psychology, University of Edinburgh, 7 George Square, Edinburgh, EH8 9JZ, UK.
• PMID: 39262091
• DOI: 10.1177/17470218241284414

Working memory is the system that supports the temporary storage and processing of information. It is generally agreed that working memory is a mental workspace, with a combination of resources operating together to maintain information in mind for potential use in thought and action. Theories typically acknowledge contributions of long-term memory to this system. One particular aspect of long-term memory, namely semantic long-term memory, can effectively supplement or 'boost' working memory performance. This may be a relatively automatic process via the semantic properties of the stimuli or more active via strategy development and implementation. However, the precise mechanisms require greater theoretical understanding. In this review of the literature, we critically discuss theoretical models of working memory and their proposed links with long-term memory. We also explore empirical research that contributes to our understanding of the ways in which semantics can support performance on both verbal and visuospatial working memory tasks, with a view to potential intervention development. This includes the possibility of training people with lower performance (e.g., older adults) to use semantics during working memory tasks. We conclude that semantics may offer an opportunity to maximise working memory performance. However, to realise this potential, more research is needed, particularly in the visuospatial domain.

Keywords: semantic long-term memory; verbal memory; visual-spatial; visuospatial memory; working memory.

PubMed Disclaimer

LinkOut - more resources

Full text sources.

• Citation Manager

NCBI Literature Resources

MeSH PMC Bookshelf Disclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

New study reveals how brain waves control working memory

MIT neuroscientists have found evidence that the brain’s ability to control what it’s thinking about relies on low-frequency brain waves known as beta rhythms.

In a memory task requiring information to be held in working memory for short periods of time, the MIT team found that the brain uses beta waves to consciously switch between different pieces of information. The findings support the researchers’ hypothesis that beta rhythms act as a gate that determines when information held in working memory is either read out or cleared out so we can think about something else.  

“The beta rhythm acts like a brake, controlling when to express information held in working memory and allow it to influence behavior,” says Mikael Lundqvist, a postdoc at MIT’s Picower Institute for Learning and Memory and the lead author of the study.

Earl Miller, the Picower Professor of Neuroscience at the Picower Institute and in the Department of Brain and Cognitive Sciences, is the senior author of the study, which appears in the Jan. 26 issue of Nature Communications .

Working in rhythm

There are millions of neurons in the brain, and each neuron produces its own electrical signals. These combined signals generate oscillations known as brain waves, which vary in frequency. In a 2016 study , Miller and Lundqvist found that gamma rhythms are associated with encoding and retrieving sensory information.

They also found that when gamma rhythms went up, beta rhythms went down, and vice versa. Previous work in their lab had shown that beta rhythms are associated with “top-down” information such as what the current goal is, how to achieve it, and what the rules of the task are.

All of this evidence led them to theorize that beta rhythms act as a control mechanism that determines what pieces of information are allowed to be read out from working memory — the brain function that allows control over conscious thought, Miller says.

“ Working memory is the sketchpad of consciousness, and it is under our control. We choose what to think about,” he says. “You choose when to clear out working memory and choose when to forget about things. You can hold things in mind and wait to make a decision until you have more information.”

To test this hypothesis, the researchers recorded brain activity from the prefrontal cortex, which is the seat of working memory, in animals trained to perform a working memory task. The animals first saw one pair of objects, for example, A followed by B. Then they were shown a different pair and had to determine if it matched the first pair. A followed by B would be a match, but not B followed by A, or A followed by C. After this entire sequence, the animals released a bar if they determined that the two sequences matched.

The researchers found that brain activity varied depending on whether the two pairs matched or not. As an animal anticipated the beginning of the second sequence, it held the memory of object A, represented by gamma waves. If the next object seen was indeed A, beta waves then went up, which the researchers believe clears object A from working memory. Gamma waves then went up again, but this time the brain switched to holding information about object B, as this was now the relevant information to determine if the sequence matched.

However, if the first object shown was not a match for A, beta waves went way up, completely clearing out working memory, because the animal already knew that the sequence as a whole could not be a match.

“T he interplay between beta and gamma acts exactly as you would expect a volitional control mechanism to act,” Miller says. “Beta is acting like a signal that gates access to working memory. It clears out working memory, and can act as a switch from one thought or item to another.”

A new model

Previous models of working memory proposed that information is held in mind by steady neuronal firing. The new study, in combination with their earlier work, supports the researchers’ new hypothesis that working memory is supported by brief episodes of spiking, which are controlled by beta rhythms. Two other recent papers from Miller’s lab offer additional evidence for beta as a cognitive control mechanism.

In a study that recently appeared in the journal Neuron , they found similar patterns of interaction between beta and gamma rhythms in a different task involving assigning patterns of dots into categories. In cases where two patterns were easy to distinguish, gamma rhythms, carrying visual information, predominated during the identification. If the distinction task was more difficult, beta rhythms, carrying information about past experience with the categories, predominated.

In a recent paper published in the Proceedings of the National Academy of Sciences , Miller’s lab found that beta waves are produced by deep layers of the prefrontal cortex, and gamma rhythms are produced by superficial layers, which process sensory information. They also found that the beta waves were controlling the interaction of the two types of rhythms.

“When you find that kind of anatomical segregation and it’s in the infrastructure where you expect it to be, that adds a lot of weight to our hypothesis,” Miller says.

The researchers are now studying whether these types of rhythms control other brain functions such as attention. They also hope to study whether the interaction of beta and gamma rhythms explains why it is so difficult to hold more than a few pieces of information in mind at once.

“Eventually we’d like to see how these rhythms explain the limited capacity of working memory, why we can only hold a few thoughts in mind simultaneously, and what happens when you exceed capacity,” Miller says. “You have to have a mechanism that compensates for the fact that you overload your working memory and make decisions on which things are more important than others.”

The research was funded by the National Institute of Mental Health, the Office of Naval Research, and the Picower JFDP Fellowship.

MIT News story

Related Articles

Study reveals how an anesthesia drug induces unconsciousness.

Consciousness

Study models how ketamine’s molecular action leads to its effects on the brain

How the brain is flexible enough for a complex world (without being thrown into chaos)

Introducing ART: A new method for testing auditory memory with circular reproduction tasks

• Original Manuscript
• Open access
• Published: 09 September 2024

Cite this article

You have full access to this open access article

• Aytaç Karabay   ORCID: orcid.org/0000-0003-3592-1531 1   na1 ,
• Rob Nijenkamp 2   na1 ,
• Anastasios Sarampalis 3 &
• Daryl Fougnie 1  

2 Altmetric

Theories of visual working memory have seen significant progress through the use of continuous reproduction tasks. However, these tasks have mainly focused on studying visual features, with limited examples existing in the auditory domain. Therefore, it is unknown to what extent newly developed memory models reflect domain-general limitations or are specific to the visual domain. To address this gap, we developed a novel methodology: the Auditory Reproduction Task (ART). This task utilizes Shepard tones, which create an infinite rising or falling tone illusion by dissecting pitch chroma and height, to create a 1–360° auditory circular space. In Experiment 1, we validated the perceptual circularity and uniformity of this auditory stimulus space. In Experiment 2, we demonstrated that auditory working memory shows similar set size effects to visual working memory—report error increased at a set size of 2 relative to 1, caused by swap errors. In Experiment 3, we tested the validity of ART by correlating reproduction errors with commonly used auditory and visual working memory tasks. Analyses revealed that ART errors were significantly correlated with performance in both auditory and visual working memory tasks, albeit with a stronger correlation observed with auditory working memory. While these experiments have only scratched the surface of the theoretical and computational constraints on auditory working memory, they provide a valuable proof of concept for ART. Further research with ART has the potential to deepen our understanding of auditory working memory, as well as to explore the extent to which existing models are tapping into domain-general constraints.

Similar content being viewed by others

The effects of visual working memory load on detection and neural processing of task-unrelated auditory stimuli

Visual memory, the long and the short of it: a review of visual working memory and long-term memory.

There Is no Theory-Free Measure of “Swaps” in Visual Working Memory Experiments

Avoid common mistakes on your manuscript.

Introduction

Theories about the nature of visual working memory have advanced considerably since the adoption of continuous reproduction tasks and mixture modeling frameworks. In continuous reproduction tasks (e.g., the delayed estimation paradigm), participants are asked to remember a feature value(s) of a stimulus (or set of stimuli) and to later reproduce this stimulus from a perceptually circular and uniform space (e.g., color; Prinzmetal et al., 1998 ; Wilken & Ma, 2004 ). In contrast to change detection and delayed report tasks (Jacobsen, 1936 ; Luck & Vogel, 1997 ), which provide discrete correct/incorrect responses, continuous reproduction tasks allow for graded memory responses. Commonly used circular continuous reproduction tasks have key benefits over dichotomous tasks (Skóra et al., 2021 ). These benefits include a more detailed perceptual or representational quality than dichotomous tasks, as they provide a raw measure of distance between the perceived and remembered stimulus; they also constitute a bias-free measure, since dichotomous tasks may involve more liberal or conservative reports (Macmillan & Creelman, 2004 ; Wilken & Ma, 2004 ). Critically, by utilizing a circular space for the report dimension, modeling frameworks are able to take a distribution of errors obtained from continuous reproduction tasks and isolate separate putative states which link to distinct cognitive mechanisms. The most well-known of these modeling frameworks is the standard mixture model (Zhang & Luck, 2008 ), which argued that a major constraint on visual memory was an upper limit on the number of items that could be stored. In this view, error distributions can be described as responses arising from the weighted mixture of two states: an “in-memory” state captured by a circular normal distribution with a width corresponding to the fuzziness of memory, and an “out-of-memory” state characterized by a uniform distribution over the possible values (since uninformed responses are random relative to the stimulus). In recent years, adaptations and revisions of this framework have led to many notable advancements in our understanding of visual working memory (for a systematic comparison of computational working memory models, see Van den Berg et al., 2014 ). While some theoretical frameworks reject the idea of separate memory states (e.g., Schurgin et al., 2020 ), the ability to construct and test generative models has led to rapid progress in our understanding.

A side effect of the focus on continuous reproduction tasks, however, has been a reduction in the scope of possible stimulus spaces in which working memory can be investigated. Continuous reproduction tasks have utilized color, orientation, location, motion, face, and shape spaces (e.g., Asplund et al., 2014 ; Bays & Husain, 2008 ; Karabay et al., 2022 ; Li et al., 2020 ; Wilken & Ma, 2004 ; Zokaei et al., 2011 ). Notably, these stimuli are all part of the visual domain. Prior to the development of mixture modeling techniques, a central question in the working memory literature was the degree to which working memory limits reflect a domain-independent limitation or whether separate limitations and properties exist for the different modalities such as vision and audition (e.g., Cowan, 1998 ; Fougnie & Marois, 2011 ; Morey et al., 2011 ; Saults & Cowan, 2007 ). Although there have been a few instances where continuous reproduction tasks were used to assess auditory working memory, the stimulus spaces utilized in these approaches lacked circularity (Clark et al., 2018 ; Hedger et al., 2018 ; Kumar et al., 2013 ; Lad et al., 2020 , 2022 ) or uniformity (Joseph et al., 2015 ), and most of them are not openly accessible (but see Lad et al., 2022 ). Because of these limitations, existing paradigms have not been adopted more generally and research on auditory working memory has not fully benefited from the computational modeling frameworks developed in recent years. These frameworks often use error responses to separate responses into distinct mechanisms. For example, some responses may reflect noisy target memories while others may reflect random responses (Zhang & Luck, 2008 ). Circular spaces simplify the separation of putative response types. Further, uniform circular spaces reduce biases and distortions that can arise in a bounded space (e.g., responses that are biased toward the center of a bounded space, Thiele et al., 2011 ). Generating an auditory stimulus space that is both circular and uniform is not as intuitively obvious as when utilizing visual features. Unlike certain visual features that lend themselves naturally to a circular space, such as color and orientation, stimuli in auditory working memory tasks have often had discrete, noncontinuous values and have varied over a bounded rather than a circular space. This makes the interpretation of putative guess responses challenging. Therefore, it is paramount to identify a stimulus set that lends itself to use in a circular continuous space that is uniform in nature.

To address this gap in the literature, we have developed an auditory continuous reproduction task that utilizes a uniform circular space consisting of Shepard tones drawn from a Shepard scale (Shepard, 1964 ). Shepard tones are complex waveforms that are known to create the auditory illusion of a tone that is infinitely rising or falling in terms of its pitch. This illusion is created by simultaneously playing back different frequency components spaced at successive octave intervals of a tonal pitch (i.e., the frequency of each component above its lowest included octave is exactly twice the frequency of the component in the octave just below), where the amplitude for each frequency component is gradually tapered off to sub-threshold levels for components that are further away from the specified fundamental frequency. For a circular reproduction task utilizing a stimulus space composed of 360 distinct Shepard tones that are logarithmically spaced in a single octave in terms of their fundamental frequencies, it is essential that the stimulus space “wraps around” a circle. Shepard tones allow for this circular wrapping, as they allow for differentiation in pitch class or pitch chroma (i.e., the specific note that is played) but are ambiguous in terms of their pitch height (i.e., the specific octave the note played is in; Deutsch, 1984 ; Deutsch et al., 2008 ; Shepard, 1964 ; Siedenburg et al., 2023 ). As a result, the transition from the higher end of the fundamental frequency range of the octave making up the Shepard scale (i.e., 360°) back to the lower end of the same octave (i.e., 1°) is imperceptible, and instead the illusion of a tone with a continuously falling or rising pitch is achieved.

The goal of this research is to provide (and make easily accessible) a proof-of-concept continuous reproduction task for auditory working memory based on Shepard tone space. Experiment 1 aims to test whether Shepard tone space meets the requirements for the Auditory Reproduction Task (ART), namely, perceptual circularity and uniformity. It is hypothesized that Shepard tone space will show perfect circularity and a uniform distribution (Deutsch, 1984 ; Deutsch et al., 2008 ; Shepard, 1964 ; Siedenburg et al., 2023 ). In Experiment 2, we introduce ART and manipulated set size that utilizes the circular Shepard tone space described above. We ask whether auditory working memory exhibits similar patterns to visual working memory, such as escalated reproduction errors with an increase in memory set size. We apply commonly used computational working memory models (e.g., standard mixture model, Zhang & Luck, 2008 ) to test underlying causes of error differences between set sizes. Moreover, we test task performance over time, assessing whether reproductions exhibit increasing precision or reveal variability with successive blocks. Next, in Experiment 3, we test the validity of ART by means of correlation with existing auditory and visual working memory paradigms. Finally, we discuss applications of ART and how it can shed light on the theoretical and computational constraints of auditory working memory, utilizing the same frameworks that have advanced our understanding of visual working memory.

Experiment 1: Validating the circularity of Shepard tones

The Shepard tone was devised to dissect the specific contributions of chroma by making the pitch height ambiguous. By increasing the chroma of Shepard tones in a sequential manner, a cyclic pattern is established where a complete octave shift is perceptually (acoustically) identical to no shift at all. This generates the widely recognized Shepard scale illusion, which creates the perceptual experience of a continuous rise or fall in pitch. The evidence that Shepard tones make up a circular perceptual space comes from direction discrimination paradigms. In such a paradigm, responses are often divided equally between perceived upward and downward shifts for the same Shepard tone when compared to another tone that is separated by half an octave (Deutsch, 1984 ; Deutsch et al., 2008 ; Shepard, 1964 ; Siedenburg et al., 2023 ). Although the perception of the Shepard tone space was found to be circular, the literature has not quantitively assessed its circularity. By using a novel circularity value, we quantitatively assessed the circularity of the Shepard tone space (Li et al., 2020 ). Further, we qualitatively investigated the uniformity of the distribution of the tones. To test these questions, we generated 15 discrete Shepard tones with fundamental frequencies making up a single octave range and asked participants to rate the similarity of the tones played. We tested the circularity of this perceptual space by applying multidimensional scaling on similarity judgments, which is done by analyzing the similarity ratings of tone pairs and then placing those stimuli in a two-dimensional space based on their perceived similarity (Shepard, 1980 ). Additionally, we qualitatively examined the distances of the tones relative to each other to further assess their uniformity.

Participants

The number of participants to assess circularity of Shepard tone space was determined by following Li et al.’s sample size ( 2020 ). As we use Bayesian tests for statistical analyses, we did not estimate sample sizes in advance, as Bayesian tests do not require sample size calculation (Berger & Wolpert, 1988 ). Participants who did not have an average similarity rating of 4 or higher on identical sound samples (e.g., tone 1 vs. tone 1) during the similarity rating phase of the experiment were excluded from further data analysis. This exclusion criterion was applied because it indicated that these participants could not sufficiently discriminate tone pairs. In total, 28 individuals participated in Experiment 1, four of which were excluded from the study due to this exclusion criterion. The final sample consisted of 24 students (11 female, 13 male) from the New York University Abu Dhabi (mean age = 20.3 years, ranging from 18 to 24 years). All participants reported having normal hearing. Participants were assigned to one of two groups depending on their participant number parity, with 12 participants assigned to group 1 and 12 participants assigned to group 2. Each of these two groups rated the subjective similarity of different stimuli sets. Participants were compensated for their participation in the experiment with vouchers worth 50 dirhams. The study was approved by the ethical committee of the New York University Abu Dhabi Psychology Department. Informed consent forms were collected prior to data collection, and the research was conducted in accordance with the Declaration of Helsinki (2008).

OpenSesame 3.3.12 (Mathôt et al., 2012 ) using the Expyriment back-end (Krause & Lindemann, 2014 ) was used for trial preparation and data collection under the Microsoft Windows 10 operating system. Participants were individually seated in sound-attenuated testing cabins about 60 cm from a 24-inch BenQ XL2411 screen. The screen resolution was set to 1024 by 768 pixels. The stimulus sounds were played through Audio-Technica ATH-M50x over-ear headphones. The playback device’s volume level was fixed at 20% of the maximum, to a comfortable, audible level. Responses were collected using a standard computer mouse.

The auditory stimulus space used in this experiment consisted of a Shepard tone complex (STC; Shepard, 1964 ). The use of the STC space proved essential for mapping the fundamental frequency of the played-back Shepard tones to a circular perceptual task, as these kinds of tone complexes are known for enabling differentiation in pitch class or pitch chroma (i.e., the specific note played) while being ambiguous in pitch height (i.e., the specific octave a note is in). The ShepardTC function (Böckmann-Barthel, 2017 ) was used in MATLAB to generate the tones used as the auditory stimulus space. Each Shepard tone is composed of different frequency components spaced at successive octave intervals of a tonal pitch within 30 and 16,000 Hz, where the amplitude for each frequency component that is further away from the specified fundamental frequency is gradually tapered off following a cos 2 bell-shaped spectral envelope until sub-threshold levels are reached. Note that this envelope procedure differs slightly from that used in the original work of Shepard ( 1964 ), where the spectral envelope was applied to the sound levels of the different frequency components rather than the amplitude. The fundamental frequencies of the tones making up the auditory stimulus space were in a single octave ranging from 278.4375 Hz to 556.875 Hz (i.e., the same pitch chroma one full octave higher). The subharmonic frequencies included a range of nine octaves between 30 and 16,000 Hz (i.e., 31.25, 62.5, 125, 250, 500, 1000, 2000, 4000 and 8000 Hz were used to create a Shepard tone with a fundamental frequency of 500 Hz). The tones were generated using a sampling frequency of 44,100 Hz. The STC space was logarithmically transformed using a base-2 (log 2 ) scale to a 1–360° circular space, such that 1° was equivalent to 279.2109 Hz and 360° was equal to 556.8750 Hz. Each increase of 1° was equal to an increase in the fundamental frequency by a factor of \(\sqrt[360]{2}\) . Fifteen discrete tones with a 24° difference were used for assessing the circularity of the perceptual space of the tones (Table  1 ).

The nature of Shepard tones allows an octave tone space to psychologically “wrap” from the end of the space to the beginning. To demonstrate this, we adopted the procedure from Li et al. ( 2020 ) to assess the circularity of our auditory space. There were a total of 163 trials, of which 10 were practice trials. Each trial had two phases: the initial judgment phase and the similarity rating phase. The experiment took approximately 60 min to complete per participant.

Initial judgment phase

Every trial started with a mouse click. After the mouse click, three speaker icons appeared on the screen, forming a triad (Fig.  1 b). First, 600 ms after the icons appeared, the tone at the top location played for 500 ms, followed by one of the two tones at the bottom locations for the same duration. Later, the tone at the top location played again for 500 ms. Finally, the tone at the remaining bottom location played for the same duration. A white circle appeared behind a speaker icon while the corresponding sound was being played. The order in which the bottom left and right sides of the triad were played was randomized and distributed equally across trials (the left tone played first in 76 trials and the right tone played first in 77 trials, or vice versa). Each tone was separated by a 600-ms inter-stimulus interval (ISI). After a 1000-ms interval after the offset of the last tone, the response screen appeared. Participants were then asked to click on one of the sounds at the two bottom locations that they perceived to be most like the sound corresponding to the top location. Data from the initial judgment phase were not used in any analyses; the purpose of this phase was to calibrate participants’ pairwise similarity ratings for the following similarity rating phase.

STC space and task demonstration. a The STC space and corresponding locations on a 360° space. Black dots on the sound circle show the anchor tones that were played to both participant groups. Blue triangles are tones used in participant group 1, and purple squares are tones used in participant group 2. b Initial judgment phase of the experiment. White disks appeared on the tone that was playing. Participants were asked to click the most similar bottom sound relative to the top sound. c Similarity rating phase. Tone pairs in the initial judgment phase were played sequentially in random order (e.g., top and left followed by top and right). Participants reported the subjective similarity of the sound samples, choosing from a Likert-type scale ranging from 0 to 5

Similarity rating phase

After a one-second inter-phase interval, a screen with two sound icons appeared, with one of them located at the top of the screen and the other located at the bottom of the screen (Fig.  1 c). The top tone was identical to the tone that was presented in the same location during the initial judgment phase. The bottom tones corresponded to the tones located on either the bottom left or right side of the triad from the previous initial judgment phase. The order of the tone pairs corresponding to the bottom location was randomized and distributed equally across trials (the left tone played first in 76 trials and the right tone played first in 77 trials, or vice versa). First, the top tone was played for 500 ms, followed by the first bottom tone that was played for the same duration with a 600-ms ISI between the tones. A 1000-ms interval was inserted between the offset of the bottom tone and the response screen. Participants were then asked to judge the similarity of the two tones using a Likert scale ranging from 0 (no similarity) to 5 (identical). After an interval of 1000 ms after participants gave their response, the same response procedure was repeated for the remaining bottom tone that was also played in the initial judgment phase. Participants were instructed to rate the similarity of the tones using the whole range of the scale (Supplementary material Fig. S1 shows that participants were able to follow this instruction successfully).

Each participant group listened to nine different tones (triangles and rectangles in Fig.  1 a), three of which were identical in both groups (black dots in Fig.  1 a). The tones that both groups listened to were used as anchor points to combine the perceptual spaces of the two groups during analysis. These anchor point tones were separated by a 120° difference on the circular auditory space (see Table  1 for the tones that were used in each participant set). Three tones were pseudo-randomly sampled in every trial during the initial judgment phase, resulting in all possible combinations of tones being sampled per participant. Each tone was compared with other tones a total of eight times (four out of eight times, one of the tones in each pair was shown at the top location of the triad during the initial judgment phase). Each tone was compared with itself twice during the experiment.

Assessing circularity

We assessed the circularity of the STC space by calculating the circularity value ( C ) of the perceptual space based on multidimensional scaling (Li et al., 2020 ), which is the ratio of the area to the perimeter of the STC space, providing a quantitative measure of its circularity. First, similarity matrices using the pairwise similarity ratings of tones were created per participant. Pairwise similarity ratings were averaged regardless of their presentation order, creating a symmetrical similarity matrix for each participant (Fig.  2 a). Next, we averaged each participant’s similarity matrix and reverse-coded them to create a single dissimilarity matrix per participant set. The averaging procedure increases the precision of the multidimensional scaling (MDS) by increasing power and reducing error (Li et al., 2020 ; Fig.  2 b). Based on the dissimilarity matrices, the subjective perceptual space of the tones was created with MDS per participant set using the mds function of the smacof package in R (Mair et al., 2022 ; Fig.  2 c). MDS is a widely used descriptive analysis based on similarity judgments. MDS maps relationships between items which can be used to infer the organization of items in a multidimensional space (Hout et al., 2013 ). In order to create a single STC perceptual space (Fig.  2 d, e), we combined the perceptual space of each participant set by applying an affine transformation to match the anchor points using the computeTransform function from the morpho package (Schlager, 2017 ). Finally, we drew a smooth curve connecting the dots on the perceptual space with a spline function (Shepard, 1980 ). C was calculated from the area and perimeter of the splined shape (orange line in Fig.  2 e) with the following function:

Illustration of the circularity analysis with fictional data. a Similarity matrix of each participant. b Group dissimilarity matrix created by averaging each group’s similarity matrix. After averaging, similarity ratings were reverse-coded. c Perceptual space of STC in each participant group. Shapes were created with multidimensional scaling of dissimilarity matrices per participant set. d Perceptual space of the second participant set after an affine transformation. e Perceptual tone space of the combined participant set. The circularity value was calculated on a splined and dashed line. Figure 2 is adapted from Li et al., ( 2020 ; Fig.  3 ) with permission from the American Psychological Association

This function captures the ratio of the area of a polygon with its perimeter. C with a value of 1 indicates a perfect circle (see Li et al., 2020 , for details). According to simulations by Li et al. ( 2020 ), a C of 0.9 is close to a perfect circle; we chose a C of 0.9 as a threshold for nearly perfect circularity (see Supplementary Fig. S2 for C value of equilateral polygons).

Statistical testing

Bayesian t -tests and ANOVAs were run with JASP 0.16.3 (JASP Team, 2022 ). Bayes factor (BF) values were interpreted according to Wetzels et al. ( 2011 ). BF 10 values of 1–3 were regarded as anecdotal evidence, 3–10 as substantial evidence, 10–30 as strong evidence, 30–100 as very strong evidence, and BF 10 values above 100 as decisive evidence in favor of the alternative hypothesis. BF 10 values of 0.33–1 were regarded as anecdotal evidence, 0.1–0.33 as substantial evidence, 0.03–0.33 as strong evidence, 0.01–0.03 as very strong evidence, and BF 10 values below 0.01 as decisive evidence in favor of the null hypothesis. Tidyverse (Wickham et al., 2019 ) and data.table R packages were used to preprocess and analyze the data. GGplot2 (Wickham, 2016 ) was used to create figures.

Results and interim discussion

Similarity ratings.

Visual inspection of the similarity matrix allowed us to assess whether the STC space wrapped to 360°. Specifically, if the STC space was perceived as circular, pairwise similarity ratings should be determined by angular rather than frequency differences. If the similarities of tone pairs followed a linear pattern, similarity ratings would not recover as the difference in fundamental frequency increased. However, pairwise similarity ratings decreased as the circular distance between tone pairs increased in both participant sets (Fig.  3 a) as well as when participant sets were combined (Fig.  3 b).

Similarity ratings of tone pairs: a per participant group and b for the combined participant sets. The combined set consists of tone pair similarity ratings for both participant sets. The x - and y -axis show the fundamental frequency of tones

Pairwise similarity ratings for each tone as a function of their angular distance were visually investigated to assess whether the perceptual STC space wrapped around a circle (Fig.  4 a). If the STC space wrapped around a circle, there should not be any sharp change in the perceptual similarity scores around the boundary of the circle (360°). Following this prediction, no such sharp change was observed with regard to the similarity ratings of the tones relative to the circular boundary (360°, red vertical lines in Fig.  4 a). The minimum pairwise rating for each tone relative to the tone it was compared to was lowest when the angular distance was the greatest (blue horizontal lines in Fig.  4 a). This means that as the angular distance between tone pairs increased, their similarity scores decreased. As a final check, we compared the similarity ratings of close (24° and 336°) and far (168° and 192°) tones relative to 360°, with the expectation that those tones with a greater circular distance would show larger differences. This was indeed what was found (Fig.  4 b). Bayesian between-subject ANOVA analysis showed decisive evidence that circular distance influences pairwise similarity ratings ( BF 10  > 1000; Fig.  4 b). Bayesian post hoc tests were conducted to compare the tone similarity ratings. Critically, there was anecdotal evidence against any differences in similarity ratings between 24° and 336°, despite the fact that 24° is the furthest from 360° in unwrapped space ( M 24°  = 3.98, SD 24°  = 0.80; M 336°  = 3.87, SD 336°  = 0.69; BF 10  = 0.39). These tests also supported general distance effects. Bayesian post hoc tests showed decisive evidence that the perceptual similarity of tones at 24° and 360° was greater than the perceptual similarity of 360° and 168° or 192° ( M 168°  = 2.32, SD 168°  = 0.84; M 192°  = 1.92, SD 192°  = 0.66; BF 10  = 622.28; BF 10  > 1000, respectively). Likewise, decisive evidence was observed when using 336° as the reference point—similarity ratings of tones at 336° and 360° were greater than 360° and 168° or 196° ( BF 10  = 591.72; BF 10  > 1000, respectively). All in all, we conclude that circular distance accounts for the Shepard tone’s perceptual similarity.

Similarity ratings for each tone and comparison of similarity ratings relative to circle boundary . a Similarity ratings as a function of angular distance for each unique tone. The red vertical lines on the panel show 360°, and the blue horizontal lines show the minimum similarity ratings. If perceptual space was not circular, then we would expect abrupt deviations from symmetry near the red vertical line. The blue color shows the fits calculated with the local polynomial regression fitting (loess) function using the formula y ~ x. Error bars represent 95% between-subject confidence intervals. Gray dots show participant data, and gray lines connect each participant’s similarity ratings. b Similarity ratings of the closest and furthest tones relative to the circle boundary, 360°. The x -axis shows corresponding angles of the tones on the STC space, and the y -axis shows similarity ratings. Error bars represent 95% between-subject confidence intervals, and the black dots show the average similarity rating. Transparent dots show each participant’s similarity ratings. Boxplot shows median and quartile values. The violin plot shows the distribution. **** Decisive evidence supporting the difference between conditions

Circularity of the Shepard tone space

We mapped the perceptual STC space for each participant set. C was 0.98 on the splined area (orange shape in Fig.  5 a) and 0.94 on the dashed shape (black dashed polygon in Fig.  5 a) in participant set 1. Similar values for C were observed for the second participant set, where C was estimated as 0.98 on the splined shape (orange shape in Fig.  5 b) and 0.95 on the dashed shape (black dashed polygon in Fig.  5 b). Combining each participant set’s perceptual space with affine transformation increased C . C was 0.99 on the splined area of the combined perceptual STC space (orange shape in Fig.  5 c) and 0.96 on the dashed area (black dashed polygon in Fig.  5 c). C with a value of 0.99 can be considered an almost perfect circle, as only an equilateral hexadecagon can achieve the same C value (Supplementary Fig. S2 ) . Furthermore, since the C value of STC’s perceptual space was above the threshold ( C all  > 0.9), it can be concluded that the perception of the STC space is indeed circular. We also investigated the perceptual space of the STC of each participant separately (Supplementary Fig. S3 ). Although there were some individual differences in perceptual STC space, the perceptual space approached perfect circularity for most participants ( C  > 0.90 for 16 participants). For the remaining participants, the perceptual space followed a circular path (with C  < 0.9), except in two cases, for which the perceptual space was not circular. Thus, the overall consistency across participants confirms the prediction that perceptual STC space would form a circle.

Subjective perceptual space of STC: a participant set 1, and b participant set 2. c Combined subjective perceptual space by using anchor points with affine transformation. Purple squares are the subjective perceptual space of participant set 1, and blue triangles are the subjective space of participant set 2. Black circles are anchor tones. C on the top right side of each panel indicates the circularity value

We assessed the uniformity of the STC space through a visual inspection for each participant group. This revealed that tone comparisons with small angular distances (24°) were perceived as more similar than tone comparisons with large angular distances (48°). Indeed, the perceptual spaces were uniform for both participant groups (Fig.  5 a, b).

Experiment 2: ART – An auditory delayed estimation paradigm

Continuous reproduction tasks extend our understanding of working memory by utilizing more detailed responses than change detection paradigms or other standard tasks. This allows researchers to formulate and test models that go beyond conceptualizing memory as a discrete, high-threshold system. This model comparison approach has been highly productive, resulting in major advances in the understanding of visual working memory. For example, theoretical models have been developed to separate responses into putative memory states, in order to understand how manipulations such as memory set size separately affect the properties of those states. While there are several competing theories and no consensus (Oberauer, 2021 ; Schurgin et al., 2020 ; Van den Berg et al., 2014 ), a method with detailed generative responses allows for richer model testing than existing approaches.

Prior to a shift toward continuous report tasks, a major thread of research on working memory was to understand the degree to which working memory systems operated in a modality-specific or modality-independent fashion. An unfortunate side effect of the aforementioned progress on visual working memory was a shift away from this important question. As the circular reproduction paradigms have become a standard way of measuring visual working memory, an equivalent paradigm in other domains has become a necessity for a more comprehensive understanding of modality-general working memory functioning. To our knowledge, only one study has used a circular reproduction task to test auditory working memory. Joseph et al. ( 2015 ) mapped English vowels on a circle using a two-dimensional formant space, and applied a mixture model to response errors obtained with a novel auditory delayed estimation paradigm. This vowel reproduction methodology utilized to gauge auditory working memory recall precision for phonemes was deemed robust. However, they observed an increase in categorical responses in the face of augmented memory load. It could be that the vowel stimuli space is inherently categorical, which can explain why responses were clustered around prototypic vowels (Iverson & Kuhl, 2000 ; Kuhl, 1991 ). Although their stimuli set wraps around a circle, the C score of the perceptual space was below the circularity threshold. As a case in point, we retrieved dimension 1 and 2 coordinates in the set size of 1 condition from the perceptual vowel space mapped by multidimensional scaling and calculated the C value. The perceptual vowel space did not resemble a perfect circle, as the C of the vowel space was 0.81 (i.e., below the threshold of a perfect circle). Overall, since the circular vowel space that was employed by Joseph et al. ( 2015 ) produced categorical-bias responses and its perceptual space cannot be considered circular, it does not satisfy the requirements of both uniformity and circularity. Further, outside of experiments conducted by Joseph and colleagues, the paradigm has not generated broad interest in studying auditory working memory.

To overcome the constraints of existing circular auditory reproduction tasks, we created a novel auditory working memory paradigm using an extension of the stimuli set of which the perceptual space was shown to be both uniform and circular in Experiment 1. Furthermore, we tested how the manipulation of set size modulates auditory reproduction performance. Manipulating the number of array items is common in working memory studies (e.g., Bays et al., 2009 ; Gorgoraptis et al., 2011 ; Zhang & Luck, 2008 ), as theoretical models can differ on how this manipulation will affect the pattern of errors. We predicted that remembering two tones would lead to higher error due to higher memory load (Bays et al., 2009 ; Gorgoraptis et al., 2011 ; Zhang & Luck, 2008 ). Further, we modeled our report data using a mixture modeling framework (Bays et al., 2009 ; Zhang & Luck, 2008 ; see Supplementary Fig. S6 for signal-detection models) to determine whether these errors arise due to an increase in guessing, less veridical on-target responses, or reporting the wrong item.

The primary goal was to provide a proof-of-concept demonstration of ART for testing formal models of auditory working memory via continuous report data. The hope was that by utilizing equivalent methodological and modeling frameworks for visual and auditory working memory, we could push the field beyond a focus on visual working memory, and toward a theoretical perspective that considers multiple modalities.

Although Bayesian tests do not require sample size calculations (Berger & Wolpert, 1988 ), we aimed to include the data of 18 participants, as the sequential analysis (Schönbrodt et al., 2017 ) of a pilot study showed that 18 participants would be sufficient for finding the intended effects. In total, 24 participants participated in the experiment, six of whom were excluded from data analysis because their average reproduction error exceeded 60° (indicating low or chance-level performance), which was our a priori allowed maximum error. Participants received 50-dirham vouchers for their participation. The final sample consisted of 18 students (11 female, 7 male) at the New York University Abu Dhabi (mean age = 19.8 years, ranging from 18 to 22 years). All participants reported normal hearing. Consent forms were collected prior to participation, and the study was conducted in accordance with the Declaration of Helsinki (2008).

The experiment was run on computers with the Windows 10 operating system, with a screen resolution set to 1280 × 1024, and using MATLAB (The MathWorks, Inc., 2020 ) with the Psychophysics Toolbox extensions (Brainard, 1997 ; Pelli, 1997 ). ATH-M50x over-ear headphones were used for sound playback, and responses were collected with a standard computer mouse.

Stimuli were generated using the same method as in Experiment 1. To construct an auditory continuous reproduction task, the full range of the stimulus space (1–360°) was used rather than discrete tones. All other details of the stimuli were identical to Experiment 1.

Following a practice block of eight trials (which participants repeated until five out of the eight trials had less than 45° error), the experiment consisted of 4 blocks of 50 trials each. Every block included an equal number of trials per set size, presented in random order. Trials were self-paced; each trial started with a mouse click, followed by 400 ms of a fixation cross (Fig.  6 a). A randomly chosen tone was presented for 750 ms, followed by a 2750-ms retention interval in the condition with a set size of 1. When the set size was 2, another randomly chosen tone was presented for 750 ms after a 1000-ms ISI following the first tone. The final tone was then followed by a 1000-ms retention interval. The total duration from the onset of the first tone until the response was identical between set size conditions, as we considered it important to try to equate (as best as possible) the duration of items in working memory across the two conditions. Participants were instructed to keep their eyes on the fixation cross that was present in the middle of the screen during the trial until the response screen. At the response screen, a circle appeared with a dot positioned at its center. Participants then dragged the dot to the circle using the mouse, locking the dot onto the circle and starting sound playback. After the dot was locked onto the circle, participants could drag the dot to any position on the circle. Changing the position of the dot on the circle, and therefore the angle that corresponded to this position on the circle, simultaneously changed the played-back tone in terms of its fundamental frequency. To prevent participants from remembering the location of the tone rather than the sound information itself, we randomly rotated the reproduction circle each trial. When moving the position of the dot on the circle, sound playback stopped and only started again after leaving the dot in a particular position. After orienting the dot to the position on the circle corresponding to their memory of the presented tone, the participants clicked the mouse to lock in its position. Since it was important for responses to be accurate instead of fast, no response time limitations were implemented. A warning stating “too fast response” appeared on the screen if the response time was less than 750 ms. For set size 2, only the first or second tone (randomly selected) was probed on a given trial. After locking in the position of the dot on the circle, trial-by-trial feedback consisting of the presented tone’s true angle and the participant’s response angle was presented for 400 ms. After the offset of the feedback, the fixation cross remained on the screen for another 400 ms until the end of the trial. The task took approximately 45 min to complete per participant.

Illustration of the experiment and measures. a Illustration of a trial. During presentation of sound 1 and sound 2, there was nothing onscreen besides the fixation; the spectrograms are for illustrative purposes only. Sound 2 (dotted box) only played on set size 2 trials. Otherwise, during this interval only the fixation remained onscreen. b Reproduction error. The white dot on the circle shows the presented tone, and the black dot represents the reported tone. The angular difference between the reported and true tones is the reproduction error. Illustration of c the standard mixture model and d swap model

Design and analysis

There were two conditions: a set size of either 1 or 2, forming a two-factorial within-subject design. Reproduction error was used as the dependent variable and was calculated using the absolute angular difference between true and reported angles (Fig.  6 b). Bayesian paired-samples t -tests were used for pairwise comparisons. The prior is described by a Cauchy distribution centered around zero with a width parameter of 0.707. The prior corresponds to a probability of 80% that the effect size lies between − 2 and 2 (Gronau et al., 2017 ).

Mixture modeling

According to the standard mixture model, response errors are best represented by a combination of two distinct states of working memory (Zhang & Luck, 2008 ; Fig.  6 c). If a stimulus is stored in working memory, the response value tends to concentrate around the true value of the stimulus, thus producing a von Mises distribution (i.e., a circular normal distribution). The precision of the memorized items is represented by the standard deviation of the von Mises distribution. As the standard deviation of the distribution decreases, responses more accurately reflect the presented items. On the other hand, if working memory fails to store any information regarding the stimulus, the response is randomly distributed and conforms to a uniform distribution. The proportion of the uniform distribution represents the guess rate formed by random responses. The standard mixture model is described by

where x is the difference between the response and target angle wrapped between –π and π, g is the guess rate, ϕ is the von Mises distribution function centered around 0, and k is the concentration parameter of the standard deviation ( σ ).

An extension of the standard mixture model was introduced with the swap model (Bays et al., 2009 ; Fig.  6 d). The swap model suggests that true random responses and responding to the wrong item (e.g., reproducing the second item when the first item is probed or vice versa) may both contribute to the guess rate. The swap model is described by

where β is the swap rate, m is the number of distractors, and x m is the difference between the response and non-target angle wrapped between –π and π.

Using MemToolbox (Suchow et al., 2013 ) and maximum likelihood estimation, the standard model was fit to each set size condition while the swap model was fit to the set size of 2 condition. We fit the models to the error measures in each set size separately and estimated model parameters per participant.

We found that performance was higher with one tone than with two. A Bayesian paired t -test showed decisive evidence in favor of lower absolute reproduction error with a set size of 1 (33.89°, SD  = 13.61°) than with a set size of 2 (42.81°, SD  = 14.70°) ( BF 10  > 1000, Fig.  7 a–c). Unavoidably, the retention interval of the second tone was shorter than that of the first tone in the set size 2 condition, which can induce a recency effect and shadow the differences between set size conditions. To test whether a recency effect contributed to the current findings, we compared the error between temporal positions (i.e., first and second tone) in the set size 2 condition. There was substantial evidence against the hypothesis of a main effect of recency on the error in the set size 2 condition ( BF 10  = 0.25; Fig.  7 d). The average error was 42.46° ( SD  = 17.33°) when the first tone was reproduced and 43.15° ( SD  = 13.62°) when the second tone was reproduced. This means that the differences between set size conditions were not affected by a recency effect.

ART performance as a function of set size and presentation order. a Plot showing degrees corresponding to the presented tone on the x -axis and reported tone on the y -axis, separated by set size. b Density distribution of the reproduction error as a function of set size. c , d Average absolute error as a function of set size ( c ) and playback order of the tones with a set size of 2 ( d ). e , f Guess rate ( e ) and precision ( f ) as estimated by the standard mixture model. g , h Guess rate ( g ) and swap rate ( h ) as estimated by swap model. i Error density distribution relative to non-target tone. Figure conventions follow Fig.  4 . **** Decisive evidence against the null hypothesis

Parameter estimations

Bayesian paired-samples t -tests showed that differences in reproduction error stemmed from a reduction in on-target responses, not precision. Non-target responses were higher ( BF 10  = 235.63; Fig.  7 e) with set size 2 (29%, SD  = 16%) than with set size 1 (18%, SD  = 13%). Contrarily, there was anecdotal evidence against the effect of set size on the precision of the reproduction errors ( BF 10  = 0.31). Average precision was 29.1° ( SD  = 13.93°) with a set size of 1 and 30.65° ( SD  = 12.05°) with a set size of 2 (Fig.  7 f). That said, changes were in the direction of larger precision errors for a set size of 2. With a larger increase in set size, we expect that precision differences would indeed be observed.

To determine whether the increase in non-target responses reflected an increase in likely guess responses or was due to participants reporting the wrong item on set size 2, response errors for set size 2 were fit with a swap model (Bays et al., 2009 ). This model decomposed the 29% non-target responses into 18% guess responses (unrelated to any stimulus) and 11% swap responses. Given that the non-target response rate for set size 1 (20%) was not much different from the guess rate estimated here ( BF 10  = 0.24; Fig.  7 g), we conclude that the major constraint on performance with two tones in our task is a difficulty in reporting the correct item. That errors in working memory tasks may reflect a difficulty in differentiating amongst stored representations is well documented in studies of visual working memory (Bae & Flombaum, 2013 ; Bays et al., 2009 ; Oberauer & Lin, 2017 ). This suggests that discrimination amongst stored signals is a modality-general constraint on working memory performance. Participants reported the incorrect item in our study more often than is typically observed in visual working memory studies (but see Gorgoraptis et al., 2011 ). This likely indicates that temporal cues are less effective than spatial cues at cuing to internally held representations (Gorgoraptis et al., 2011 ). Future research could determine whether spatially separating tones would improve performance by reducing confusability at response.

Error across blocks

In examining performance stability across blocks of trials in ART, a Bayesian within-subject ANOVA was conducted with block number and set size as the independent variables and absolute error in ART as the dependent variable (Fig.  8 ). For the sake of simplicity, we reported only Bayesian-inclusion factors ( BF 10-inclusion ) across all-models. The analysis revealed decisive evidence favoring the main effect of set size ( BF 10-inclusion  > 1000), substantial evidence against the main effect of block progression ( BF 10-inclusion  = 0.19), and strong evidence against the interaction of block progression and set size on absolute errors ( BF 10-inclusion  = 0.08). The evidence against the effect of block progression on performance indicates that participants’ performance remained stable throughout the experiment. This suggests that time-on-task factors such as learning, fatigue, or adaptation did not significantly influence the ART errors over the course of the experiment, pointing to the robustness of the task design and the reliability of the auditory reproduction measure in assessing participants’ auditory working memory performance. Furthermore, this finding suggests that if a researcher is only interested in an estimate of performance, and not modeling the reproduction errors, a shorter experiment of about 25 trials per condition would be sufficient, which can decrease the duration of the experiment substantially.

ART errors as a function of set size and block index. Lines are regression fits per set size condition, and shaded areas are the error of the fit. Conventions and other details of the figure follow Fig.  4

Experiment 3: Validity of ART

To assess the validity of ART, we correlated ART errors with accuracy in delayed-match-to-sample tasks (DMST) across both auditory and visual domains. DMST, a widely utilized task for evaluating working memory, requires participants to maintain a set of stimuli followed by a delay period. After this delay, participants are asked to determine whether a probe stimulus was part of the initial stimulus set (see Daniel et al., 2016 , for a review). Bayesian evidence for a correlation between DMST accuracy and ART errors would provide evidence for convergent validity that ART indeed measures working memory. In addition, under the assumption that there are modality-specific components to working memory tasks (Baddeley, 2003 ; Fougnie & Marois, 2011 ; Fougnie et al., 2015 ), if the correlation between auditory DMST and ART is greater than visual DMST and ART, this finding would provide some evidence that ART draws on auditory working memory. To investigate the validity of the task, participants were administered both auditory and visual versions of the DMST, in addition to the ART, with a set size of 3. We increased the set size to avoid possible ceiling effects in all tasks.

In total, 11 participants from Experiment 2 and 13 new participants participated in the experiment. Participants received 50-dirham vouchers for their participation. The final sample consisted of 24 students (10 female, 14 male) at the New York University Abu Dhabi (mean age = 20.9 years, ranging from 18 to 25 years). All participants reported normal hearing, (corrected-to-)normal vision, and normal color vision. Consent forms were collected prior to participation, and the study was conducted in accordance with the Declaration of Helsinki (2008).

The experiment was run on computers with the Windows 10 operating system, with a screen resolution set to 1280 × 1024 at 16-bit color depth, and using MATLAB (The MathWorks, Inc., 2020 ) with the Psychophysics Toolbox extensions (Brainard, 1997 ; Pelli, 1997 ) for ART and OpenSesame 3.3.12 (Mathôt et al., 2012 ) using the Expyriment back-end (Krause & Lindemann, 2014 ) for auditory and visual DMST. ATH-M50x over-ear headphones were used for sound playback, and responses were collected with a standard computer keyboard and mouse.

ART: All details of the stimuli were identical to Experiment 2.

Auditory DMST: Twelve tones were spaced logarithmically using base-2 in a frequency range from 1000 to 2000 Hz. The tone increase was equal to an increase in frequency by a factor of \(\sqrt[12]{2}\) . Following tone generation, we equalized the root mean square of each of the tones to equalize their amplitude to ensure that low-level features of the sound stimuli were identical. The tones were generated using a sampling frequency of 44,100 Hz.

Visual DMST: Twelve colors with an angular difference of 30° were retrieved from the CIELAB color circle space (Suchow et al., 2013 ). Colors were presented in a circle with a radius of 0.65° visual angle placed at the center of the screen.

Three different tasks were used in the experiment: ART, auditory DMST, and visual DMST. The task order was randomized and counterbalanced between participants. The total duration of the experiment was one hour.

ART: The task was identical to Experiment 2 except for the following changes (Fig.  6 a). A set size of 3 was used to increase the working memory load to avoid ceiling effects, which may confound the correlation analysis with auditory and visual DMST. There were 80 experimental trials.

DMST: There were eight practice trials in both auditory and visual DMST followed by 48 experimental trials. Three random tones (auditory DMST) or colors (visual DMST) without replacement were chosen for each trial, and each was presented for 750 ms with 1000 ms of ISI between memory items (Fig.  9 ). There were two types of trials (each equally likely); in the probe-present condition, the probe was included in the sample. In the probe-absent condition, the probe was not included in the sample. In the probe-present condition, the probe was randomly chosen from the first to the third item (equally likely) during the memory phase. There was no response time limitation, and participants were asked to stress accuracy. The fixation color changed after the memory phase to inform participants that the currently presented sound or color was the probe. Participants reported the presence or absence of the probe by pressing “m” or “c” on a standard keyboard. A happy smiley (correct) or unhappy smiley (incorrect) appeared for 200 ms as feedback.

Illustration of the a auditory and b visual DMST. a During tone presentations, there was nothing onscreen besides the fixation; the spectrograms are for illustrative purposes only

Convergent validity of the ART was supported by Bayesian evidence of correlations with both auditory and visual DMST (Fig.  10 ). Decisive evidence in favor of a strong correlation between ART error and accuracy of the auditory DMST ( r  =  − 0.73, BF 10  = 513.91) was observed, indicating that as the error in ART increased, the working memory accuracy for auditory stimuli decreased. Very strong evidence in favor of a moderate negative correlation ( r  =  − 0.62, BF 10  = 32.09) between ART and accuracy in the visual DMST was observed, though to a lesser extent. The finding of a particularly large correlation between ART and an auditory working memory task suggests that ART draws on similar resources/mechanisms as auditory working memory. Lastly, we investigated test − retest reliability by correlating average ART error in Experiment 2 and Experiment 3. Since the number of participants who participated in both Experiments 2 and 3 was limited, we ran a Bayesian Kendall’s tau rank correlation test. The test results showed strong evidence favoring robust test–retest reliability ( r  = 0.78, BF 10  = 50.92), indicating that ART performance was stable over time, even when the test sessions were separated over weeks.

Scatter plot illustrating the correlation between ART errors and DMST accuracy in auditory and visual modalities. The x -axis represents the absolute ART error, and the y -axis represents the percentage accuracy of the DMST. Red points and the corresponding regression line represent the relationship between ART errors and auditory DMST accuracy, while blue points and the corresponding regression line represent visual DMST accuracy. Each point denotes a single participant’s performance. The shaded areas around the regression lines indicate the 95% confidence intervals for the estimated relationships. *** Very strong evidence and **** decisive evidence against the null hypothesis

General discussion

In recent years, remarkable progress has been made in understanding visual working memory. One of the catalysts of this progress was the implementation of modeling frameworks on continuous reproduction tasks which attempt to reverse-engineer the properties of memory representations that could generate the reproduction errors. While there is still considerable debate over which theoretical model is correct and the degree to which responses arise from separate, and isolatable states (see Oberauer, 2021 ; Schurgin et al., 2020 ), there is no debating the influence these models have had on the field of working memory and how this method has shifted the questions that the field is focused on. Unfortunately, the theoretical model framework has narrowed the scope of stimuli used to test working memory. Although there are a few examples of reproduction tasks in the auditory working memory domain, they currently do not mimic commonly used visual tasks due to a lack of circularity and/or uniformity. In this paper, we set out to create a proof-of-concept Auditory Reproduction Task (ART) and test whether the commonly used computational theories of visual working memory are also applicable to auditory working memory. To our knowledge, ART is the first auditory working memory paradigm based on continuous circular reports that utilizes a stimulus space that is free of categorical representations, and is perceptually circular and uniform. In Experiment 1, we first validated the circularity and uniformity of the perceptual space of the Shepard tones used as the auditory stimuli. After verifying this stimulus set, ART was tested within a working memory paradigm with a set size manipulation in Experiment 2. All pre-registered hypotheses were confirmed for both experiments. Error distributions were comparable to those observed in visual continuous reproduction paradigms, with the average error increasing as a function of set size. Furthermore, standard mixture and swap models fit reasonably well on the error distributions obtained through ART (see Supplementary material Fig. S7 for posterior fits per participant). The modeling outcomes suggested that swap errors can explain the worse performance observed with a set size of 2 as compared to a set size of 1, suggesting that misreporting is a modality-general constraint on working memory. In our final study, we found that performance in the ART correlated with delayed-match-to-sample working memory tasks, with a particularly strong correlation between ART and the auditory version of the task, consistent with the argument that ART draws on auditory working memory.

Perceptual space of Shepard tones

The Shepard tone was created to explore the role of chroma in pitch perception. The perceptual discrimination of Shepard tones follows circular distance rather than linear distance, as was shown with direction discrimination paradigms (Deutsch, 1984 ; Deutsch et al., 2008 ; Shepard, 1964 ; Siedenburg et al., 2023 ). Extending previous work, we quantified the circularity of the Shepard tone space in Experiment 1. To our knowledge, we are the first to quantify the circularity of the Shepard tones’ perceptual space. We found that it resembled a perfect circle with a circularity score of 0.99 in Experiment 1 (to put this into perspective: achieving a circularity score of this magnitude requires a 19-sided equilateral polygon; Supplementary Fig. S2 ). Further, by applying multidimensional scaling to each participant, we observed a striking consistency in the circularity of the perceptual space across individuals. Almost all participants’ perceptual spaces of STC adhered closely to a circular configuration (Supplementary Fig. S3 ). To confirm our findings, we calculated similarity coefficients of target–response pairs from ART following the methodology of Joseph et al. ( 2015 ), and assessed their circularity (Supplementary Fig. S4a ). The results were even more impressive, as the circularity score was increased to 1.0 when the target and response pairs were binned into 18 equal intervals. Building on this, we revisited the individual perceptual spaces in Experiment 2 and observed improved consistency across participants (Fig. S4b). This enhanced consistency might be attributable to the continuous nature of the target–response pairs used in Experiment 2, in contrast to the subset of tones employed in Experiment 1.

Besides the shape of the perceptual space, an equally important aspect of continuous reproduction paradigms is the uniformity of the tones making up this space. Although the uniformity of the STC space was not tested quantitively, the distances between the individual tones making up the perceptual space for each participant group seemed to be uniformly distributed. As a result, tones with smaller angular distances were perceived as more similar relative to distant tone pairs. Similar to the perceptual space, supplementary analysis (Supplementary Figs. S4 and S5 ) resembled a perfectly distributed uniform space of the tones when the target and response pairs of Experiment 2 were subjected to multidimensional scaling. As a final check of uniformity, we analyzed the target–response angle pairs for each participant per set size condition in Experiment 2. A visual inspection of these pairs indicated an absence of heavy clustering, suggesting that participants did not exhibit a strong bias toward any specific region of the stimulus space. This further supports the uniform distribution of responses across the circular stimulus space (Supplementary Fig. S5). Last but not least, no categorical representation of the responses was observed, as responses were not clustered around any stimulus value. Therefore, we concluded that the Shepard tones are uniformly distributed on a circle.

Auditory reproduction task

To date, the majority of auditory working memory tasks have generally utilized discrete stimuli, limiting responses to binary levels (e.g., Fougnie & Marois, 2011 ; Saults & Cowan, 2007 ). Although there were a few examples of continuous reproduction paradigms in the auditory modality, they did not become widely adopted as in visual working memory. The reason for this could be the existing paradigms’ limitations or the methodologies’ availability. While both the vowel reproduction (Joseph et al., 2015 ) and ART tasks have their respective strengths, ART has certain advantages that make it a more suitable option for testing auditory working memory in a continuous circular space. First and most importantly, ART is based on a novel stimulus set that has been validated to be both circular and uniformly distributed. Second, using a novel stimulus set can prevent possible long-term memory interference in working memory paradigms from previous exposure (Blalock, 2015 ; Jackson & Raymond, 2008 ), as ART errors are bias-free without categorical responses (e.g., categorical bias in vowel reproduction tasks, Joseph et al., 2015 ; and color reproduction tasks, Bae et al., 2014 ). Furthermore, using a novel auditory stimulus set should ensure that neural representations reflect auditory information rather than conceptual or semantic information (see Heinen et al., 2023 , for a review). Additionally, ART exhibited a correlation with delayed-match-to-sample working memory tasks, showing a strong association in particular with the auditory version of these tasks, which aligns with the view that ART taps into auditory working memory capabilities. Additionally, ART correlated with delayed-match-to-sample working memory tasks, with a particularly strong correlation between ART and the auditory version of the task, consistent with the argument that ART draws on auditory working memory. Last but not least, ART has proven to be a robust method, demonstrating stable performance across trials and sessions. This stability and specificity suggest that ART could be a useful tool for exploring the differences between auditory and visual memory processing in cognitive research.

Auditory and visual working memory reproduction compared

One of the central questions motivating ART is whether the theoretical models developed in the visual working memory literature reflect domain-general or domain-specific constraints. While our Experiment 2 reflects only a start in answering this question, the findings do suggest some common constraints on working memory, as well as some possible differences between the auditory and visual domains. Participants’ performance on ART was broadly similar to that observed in the visual working memory literature. As in visual working memory paradigms, errors increased for set size 2 relative to set size 1. When the source of error was examined via a mixture modeling analysis that separates changes in response error as arising due to a loss of precision, an increase in random responses, or swap errors, we found a significant source of response errors at set size 2, but no difference in memory precision or random responses. Consistent with this, several studies in the visual working memory literature highlight that differentiating between stored information can result in correspondence errors or difficulty in reporting the correct stimulus (Bae & Flombaum, 2013 ; Bays et al., 2009 ; Oberauer & Lin, 2017 ). That similar errors arise for visual and auditory working memory suggests that this may be a shared constraint on performance.

In addition to highlighting some similarities, the present work also suggests some potential areas of divergence between visual and auditory reproduction tasks. The average error (and precision parameter) was relatively high compared to average errors reported in the visual working memory literature (e.g., Gorgoraptis et al., 2011 ; Karabay et al., 2022 ; Zhang & Luck, 2008 ). This difference has several potential explanations, including a lower capacity of auditory working memory (Fougnie & Marois, 2011 ; Lehnert & Zimmer, 2006 ), a lower representational quality of auditory memory (Gloede & Gregg, 2019 ), differences in the similarity/discriminability of the stimulus space (see Schurgin et al., 2020 ), greater confusability amongst auditory items due to the lack of spatial location as an efficient marker, or differences in responses due to constant and changing auditory input. Further, visual working memory studies typically find decreased memory precision with larger set size (greater standard deviation of error distributions), even if the increase in set size is still lower than the presumed item capacity (e.g., Zhang & Luck, 2008 ; but see Bae & Flombaum, 2013 ). Our estimates of memory precision for set size 1 and 2 were equivalent. Does this mean that, unlike visual working memory, auditory working memory has no difference in representational quality? There are reasons to be skeptical. Precision differences in visual working memory tend to be relatively small, and our precision estimates were relatively imprecise, which would lead to noisier estimates. Future research is necessary to address this question. Finally, our swap error estimates were larger at set size 2 than is found for most studies on visual working memory. However, we believe this reflects the use of a temporal order cue rather than differences in stimulus modality. Indeed, even larger swap rates were found within the visual domain when temporal order was used as the cue (Gorgoraptis et al., 2011 ). Practically, researchers will need to be careful to minimize confusability when increasing set size beyond two items.

Critically, the goal of analyzing our data was not to adjudicate between competing theoretical models, but to show that auditory working memory can be part of this conversation. Therefore, we were focused on interpreting our findings in terms of the more dominant models in the literature. Notably, more recent models have rejected the idea of separate, putative states. According to the target confusability competition model, error distributions can be explained by a single memory strength parameter (Schurgin et al., 2020 ). When applied to our results, the memory strength parameter decreased at set size 2 compared to set size 1 (Supplementary Fig. S6 ). Regardless of which theoretical model one prefers, the ability to generalize models into another sensory domain will be of value.

Future directions

ART generates several testable research questions. For example, there are different types of distortions that have been identified in visual working memory, such as ensemble integrations, central tendency or repulsion, and attractions (Brady & Alvarez, 2011 ; Chunharas et al., 2022 ; Hollingworth, 1910 ). We do not know, however, whether these distortions are specific to visual working memory, or whether they represent a modality-independent function of working memory. If these working memory distortions are not specific to vision alone, it would be expected that similar distortions exist for auditory working memory.

We aimed to fill a gap in the literature on auditory continuous reproduction tasks with a novel methodology: ART. After confirming the circularity and uniformity of the Shepard tone space, we demonstrated the applicability of this task to the common theoretical working memory models in the literature. We think that ART has potential to benefit the field by allowing theoretical and computational working memory models built on visual continuous reproduction tasks to be tested in auditory modality, enabling a shift in the working memory field from the visual modality to a more general modality-independent understanding of working memory. In sum, ART provides a promising avenue for future research in this field and has the potential to contribute significantly to our understanding of cognitive processes beyond working memory.

Data Availability

All data including experimental tasks are publicly accessible on the Open Science Framework with the identifier WF5UV (osf.io/wf5uv/).

Code availability

Analysis scripts are publicly accessible on the Open Science Framework with the identifier WF5UV (osf.io/wf5uv/).

Asplund, C. L., Fougnie, D., Zughni, S., Martin, J. W., & Marois, R. (2014). The attentional blink reveals the probabilistic nature of discrete conscious perception. Psychological Science, 25 (3), 824–831. https://doi.org/10.1177/0956797613513810

Article   PubMed   Google Scholar  

Baddeley, A. (2003). Working memory: Looking back and looking forward. Nature Reviews Neuroscience, 4 (10), 829–839. https://doi.org/10.1038/nrn1201

Bae, G. Y., & Flombaum, J. I. (2013). Two Items Remembered as Precisely as One. Psychological Science, 24 (10), 2038–2047. https://doi.org/10.1177/0956797613484938

Bae, G.-Y., Olkkonen, M., Allred, S. R., Wilson, C., & Flombaum, J. I. (2014). Stimulus-specific variability in color working memory with delayed estimation. Journal of Vision, 14 (4), 7–7. https://doi.org/10.1167/14.4.7

Bays, P. M., Catalao, R. F. G., & Husain, M. (2009). The precision of visual working memory is set by allocation of a shared resource. Journal of Vision, 9 (10), 7–7. https://doi.org/10.1167/9.10.7

Article   Google Scholar  

Bays, P. M., & Husain, M. (2008). Dynamic shifts of limited working memory resources in human vision. Science, 321 (5890), 851–854. https://doi.org/10.1126/science.1158023

Article   PubMed   PubMed Central   Google Scholar  

Berger, J. O., & Wolpert, R. L. (1988). The Likelihood Principle (2. edition). Institute of Mathematical Statistics.

Blalock, L. D. (2015). Stimulus familiarity improves consolidation of visual working memory representations. Attention, Perception, & Psychophysics, 77 (4), 1143–1158. https://doi.org/10.3758/s13414-014-0823-z

Böckmann-Barthel, M. (2017). ShepardTC (1.0). MATLAB Central File Exchange. https://www.mathworks.com/matlabcentral/fileexchange/65443-shepardtc . Accessed 1 Nov 2022.

Brady, T. F., & Alvarez, G. A. (2011). Hierarchical encoding in visual working memory. Psychological Science, 22 (3), 384–392. https://doi.org/10.1177/0956797610397956

Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision, 10 (4), 433–436. https://doi.org/10.1163/156856897x00357

Chunharas, C., Rademaker, R. L., Brady, T. F., & Serences, J. T. (2022). An adaptive perspective on visual working memory distortions. Journal of Experimental Psychology: General, 151 (10), 2300–2323. https://doi.org/10.1037/xge0001191

Clark, K. M., Hardman, K. O., Schachtman, T. R., Saults, J. S., Glass, B. A., & Cowan, N. (2018). Tone series and the nature of working memory capacity development. Developmental Psychology, 54 (4), 663–676. https://doi.org/10.1037/dev0000466

Cowan, N. (1998). Visual and auditory working memory capacity. Trends in Cognitive Sciences, 2 (3), 77. https://doi.org/10.1016/s1364-6613(98)01144-9

Daniel, T. A., Katz, J. S., & Robinson, J. L. (2016). Delayed match-to-sample in working memory: A BrainMap meta-analysis. Biological Psychology, 120 , 10–20. https://doi.org/10.1016/j.biopsycho.2016.07.015

Deutsch, D. (1984). A musical paradox. Music Perception, 3 (3), 275–280. https://doi.org/10.2307/40285337

Deutsch, D., Dooley, K., & Henthorn, T. (2008). Pitch circularity from tones comprising full harmonic series. The Journal of the Acoustical Society of America, 124 (1), 589–597. https://doi.org/10.1121/1.2931957

Fougnie, D., & Marois, R. (2011). What limits working memory capacity? Evidence for modality-specific sources to the simultaneous storage of visual and auditory arrays. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37 (6), 1329–1341. https://doi.org/10.1037/a0024834

Fougnie, D., Zughni, S., Godwin, D., & Marois, R. (2015). Working memory storage is intrinsically domain specific. Journal of Experimental Psychology: General, 144 (1), 30–47. https://doi.org/10.1037/a0038211

Gloede, M. E., & Gregg, M. K. (2019). The fidelity of visual and auditory memory. Psychonomic Bulletin & Review, 26 (4), 1325–1332. https://doi.org/10.3758/s13423-019-01597-7

Gorgoraptis, N., Catalao, R. F. G., Bays, P. M., & Husain, M. (2011). Dynamic Updating of Working Memory Resources for Visual Objects. Journal of Neuroscience, 31 (23), 8502–8511. https://doi.org/10.1523/jneurosci.0208-11.2011

Gronau, Q. F., Van Erp, S., Heck, D. W., Cesario, J., Jonas, K. J., & Wagenmakers, E.-J. (2017). A Bayesian model-averaged meta-analysis of the power pose effect with informed and default priors: The case of felt power. Comprehensive Results in Social Psychology, 2 (1), 123–138. https://doi.org/10.1080/23743603.2017.1326760

Heinen, R., Bierbrauer, A., Wolf, O. T., & Axmacher, N. (2023). Representational formats of human memory traces. Brain Structure and Function . https://doi.org/10.1007/s00429-023-02636-9

Hollingworth, H. L. (1910). The central tendency of judgment. The Journal of Philosophy, Psychology and Scientific Methods, 7 (17), 461. https://doi.org/10.2307/2012819

Hout, M. C., Papesh, M. H., & Goldinger, S. D. (2013). Multidimensional scaling. Wiley Interdisciplinary Reviews: Cognitive Science, 4 (1), 93–103. Portico. https://doi.org/10.1002/wcs.1203

Iverson, P., & Kuhl, P. K. (2000). Perceptual magnet and phoneme boundary effects in speech perception: Do they arise from a common mechanism? Perception & Psychophysics, 62 (4), 874–886. https://doi.org/10.3758/bf03206929

Jackson, M. C., & Raymond, J. E. (2008). Familiarity enhances visual working memory for faces. Journal of Experimental Psychology: Human Perception and Performance, 34 (3), 556–568. https://doi.org/10.1037/0096-1523.34.3.556

Jacobsen, C. F. (1936). Studies of cerebral function in primates. I. The functions of the frontal association areas in monkeys. Comparative Psychology Monographs, 13 , 3 , 1–60.

JASP Team. (2022). JASP (Version 0.16.3).

Joseph, S., Iverson, P., Manohar, S., Fox, Z., Scott, S. K., & Husain, M. (2015). Precision of working memory for speech sounds. Quarterly Journal of Experimental Psychology, 68 (10), 2022–2040. https://doi.org/10.1080/17470218.2014.1002799

Karabay, A., Wilhelm, S. A., de Jong, J., Wang, J., Martens, S., & Akyürek, E. G. (2022). Two faces of perceptual awareness during the attentional blink: Gradual and discrete. Journal of Experimental Psychology: General, 151 (7), 1520–1541. https://doi.org/10.1037/xge0001156

Krause, F., & Lindemann, O. (2014). Expyriment: A Python library for cognitive and neuroscientific experiments. Behavior Research Methods, 46 (2), 416–428. https://doi.org/10.3758/s13428-013-0390-6

Kuhl, P. K. (1991). Human adults and human infants show a “perceptual magnet effect” for the prototypes of speech categories, monkeys do not. Perception & Psychophysics, 50 (2), 93–107. https://doi.org/10.3758/bf03212211

Kumar, S., Joseph, S., Pearson, B., Teki, S., Fox, Z. V., Griffiths, T. D., & Husain, M. (2013). Resource allocation and prioritization in auditory working memory. Cognitive Neuroscience, 4 (1), 12–20. https://doi.org/10.1080/17588928.2012.716416

Lad, M., Billig, A. J., Kumar, S., & Griffiths, T. D. (2022). A specific relationship between musical sophistication and auditory working memory. Scientific Reports, 12 (1). https://doi.org/10.1038/s41598-022-07568-8

Lad, M., Holmes, E., Chu, A., & Griffiths, T. D. (2020). Speech-in-noise detection is related to auditory working memory precision for frequency. Scientific Reports, 10 (1). https://doi.org/10.1038/s41598-020-70952-9

Lehnert, G., & Zimmer, H. D. (2006). Auditory and visual spatial working memory. Memory & Cognition, 34 (5), 1080–1090. https://doi.org/10.3758/bf03193254

Li, A. Y., Liang, J. C., Lee, A. C. H., & Barense, M. D. (2020). The validated circular shape space: Quantifying the visual similarity of shape. Journal of Experimental Psychology: General, 149 (5), 949–966. https://doi.org/10.1037/xge0000693

Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390 (6657), 279–281. https://doi.org/10.1038/36846

Macmillan, N. A., & Creelman, C. D. (2004). Detection theory: A user’s guide . Psychology Press.

Book   Google Scholar  

Mair, P., Groenen, P. J. F., & de Leeuw, J. (2022). More on multidimensional scaling in R: smacof version 2. Journal of Statistical Software, 102 (10), 1–47. https://doi.org/10.18637/jss.v102.i10 .

Mathôt, S., Schreij, D., & Theeuwes, J. (2012). OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44 (2), 314–324. https://doi.org/10.3758/s13428-011-0168-7

Morey, C. C., Cowan, N., Morey, R. D., & Rouder, J. N. (2011). Flexible attention allocation to visual and auditory working memory tasks: Manipulating reward induces a trade-off. Attention, Perception, & Psychophysics, 73 (2), 458–472. https://doi.org/10.3758/s13414-010-0031-4

Oberauer, K. (2021). Measurement models for visual working memory—A factorial model comparison. Psychological Review . https://doi.org/10.1037/rev0000328

Oberauer, K., & Lin, H.-Y. (2017). An interference model of visual working memory. Psychological Review, 124 (1), 21–59. https://doi.org/10.1037/rev0000044

Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial Vision, 10 , 437–442. https://doi.org/10.1163/156856897x00366

Prinzmetal, W., Amiri, H., Allen, K., & Edwards, T. (1998). Phenomenology of attention: I. Color, location, orientation, and spatial frequency. Journal of Experimental Psychology: Human Perception and Performance, 24 (1), 261–282. https://doi.org/10.1037/0096-1523.24.1.261

The Math Works, Inc. (2020). MATLAB (Version 2020b) [Computer software]. https://www.mathworks.com/

Thiele, J. E., Pratte, M. S., & Rouder, J. N. (2011). On perfect working-memory performance with large numbers of items. Psychonomic Bulletin & Review, 18 (5), 958–963. https://doi.org/10.3758/s13423-011-0108-7

Saults, J. S., & Cowan, N. (2007). A central capacity limit to the simultaneous storage of visual and auditory arrays in working memory. Journal of Experimental Psychology: General, 136 (4), 663–684. https://doi.org/10.1037/0096-3445.136.4.663

Schlager, S. (2017). Morpho and Rvcg – Shape Analysis in R. In Zheng, G., Li, S., & Szekely, G. (Eds.), Statistical Shape and Deformation Analysis , 217–256. Elsevier Gezondheidszorg. https://doi.org/10.1016/c2015-0-06799-5

Schönbrodt, F. D., Wagenmakers, E.-J., Zehetleitner, M., & Perugini, M. (2017). Sequential hypothesis testing with Bayes factors: Efficiently testing mean differences. Psychological Methods, 22 (2), 322–339. https://doi.org/10.1037/met0000061

Schurgin, M. W., Wixted, J. T., & Brady, T. F. (2020). Psychophysical scaling reveals a unified theory of visual memory strength. Nature Human Behaviour, 4 (11), 1156–1172. https://doi.org/10.1038/s41562-020-00938-0

Siedenburg, K., Graves, J., & Pressnitzer, D. (2023). A unitary model of auditory frequency change perception. PLOS Computational Biology, 19 (1), e1010307. https://doi.org/10.1371/journal.pcbi.1010307

Shepard, R. N. (1964). Circularity in judgments of relative pitch. The Journal of the Acoustical Society of America, 36 (12), 2346–2353. https://doi.org/10.1121/1.1919362

Shepard, R. N. (1980). Multidimensional scaling, tree-fitting, and clustering. Science, 210 , 390–398. https://doi.org/10.1126/science.210.4468.390

Skóra, Z., Ciupińska, K., Del Pin, S. H., Overgaard, M., & Wierzchoń, M. (2021). Investigating the validity of the Perceptual Awareness Scale – The effect of task-related difficulty on subjective rating. Consciousness and Cognition, 95 , 103197. https://doi.org/10.1016/j.concog.2021.103197

Suchow, J. W., Brady, T. F., Fougnie, D., & Alvarez, G. A. (2013). Modeling visual working memory with the MemToolbox.  Journal of Vision ,  13 (10):9, 1–8. https://doi.org/10.1167/13.10.9 .

Van den Berg, R., Awh, E., & Ma, W. J. (2014). Factorial comparison of working memory models. Psychological Review, 121 (1), 124–149. https://doi.org/10.1037/a0035234

Van Hedger, S. C., Heald, S. L., & Nusbaum, H. C. (2018). Long-term pitch memory for music recordings is related to auditory working memory precision. Quarterly Journal of Experimental Psychology, 71 (4), 879–891. https://doi.org/10.1080/17470218.2017.1307427

Wetzels, R., Matzke, D., Lee, M. D., Rouder, J. N., Iverson, G. J., & Wagenmakers, E.-J. (2011). Statistical evidence in experimental psychology. Perspectives on Psychological Science, 6 (3), 291–298. https://doi.org/10.1177/1745691611406923

Wickham, H. (2016). Ggplot2: Elegant graphics for data analysis . Springer.

Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L., François, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T., Miller, E., Bache, S., Müller, K., Ooms, J., Robinson, D., Seidel, D., Spinu, V., … Yutani, H. (2019). Welcome to the Tidyverse. Journal of Open Source Software, 4 (43), 1686. https://doi.org/10.21105/joss.01686

Wilken, P., & Ma, W. J. (2004). A detection theory account of change detection. Journal of Vision, 4 (12), 11. https://doi.org/10.1167/4.12.11

Zhang, W., & Luck, S. J. (2008). Discrete fixed-resolution representations in visual working memory. Nature, 453 (7192), 233–235. https://doi.org/10.1038/nature06860

Zokaei, N., Gorgoraptis, N., Bahrami, B., Bays, P. M., & Husain, M. (2011). Precision of working memory for visual motion sequences and transparent motion surfaces. Journal of Vision, 11 (14), 2–2. https://doi.org/10.1167/11.14.2

Download references

The study was partially supported from NYUAD Center for Brain and Health, funded by Tamkeen under NYU Abu Dhabi Research Institute grant (CG012).

Author information

Aytaç Karabay and Rob Nijenkamp contributed to the project equally.

Authors and Affiliations

Program in Psychology, New York University Abu Dhabi, Abu Dhabi, United Arab Emirates

Aytaç Karabay & Daryl Fougnie

Center for Information Technology, University of Groningen, Groningen, The Netherlands

Rob Nijenkamp

Department of Psychology, Experimental Psychology, University of Groningen, Groningen, The Netherlands

Anastasios Sarampalis

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Aytaç Karabay .

Ethics declarations

Conflicts of interest.

The authors declare no conflict of interest.

Ethics approval

The study was approved by the ethical committee of the New York University Abu Dhabi Psychology Department.

Consent to participate

Participants signed informed consent forms prior to data collection.

Consent for publication

All authors consent to publication.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Practices Statement

Experiment 3 was not pre-registered, but the stimulus, data, and analysis scripts are publicly accessible on the Open Science Framework with the identifier WF5UV (osf.io/wf5uv/). Experiments 1 and 2 were pre-registered on the same Open Science Framework repository. This public preregistration comprised the design, hypotheses, analytical approach, randomizations, and experimental programs. All analyses on the main text in Experiments 1 and 2 were pre-registered. The data collected for this study, the scripts used to generate the stimuli and the task, and the scripts used for analysis have since been uploaded to the same Open Science Framework repository.

Supplementary information

Below is the link to the electronic supplementary material.

Supplementary file1 (PDF 856 KB)

Rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Karabay, A., Nijenkamp, R., Sarampalis, A. et al. Introducing ART: A new method for testing auditory memory with circular reproduction tasks. Behav Res (2024). https://doi.org/10.3758/s13428-024-02477-2

Download citation

Accepted : 10 July 2024

Published : 09 September 2024

DOI : https://doi.org/10.3758/s13428-024-02477-2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Auditory delayed estimation paradigm
• Auditory working memory
• Shepard tone complex
• Find a journal
• Publish with us
• Track your research

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• HHS Author Manuscripts

Working Memory Underpins Cognitive Development, Learning, and Education

Working memory is the retention of a small amount of information in a readily accessible form. It facilitates planning, comprehension, reasoning, and problem-solving. I examine the historical roots and conceptual development of the concept and the theoretical and practical implications of current debates about working memory mechanisms. Then I explore the nature of cognitive developmental improvements in working memory, the role of working memory in learning, and some potential implications of working memory and its development for the education of children and adults. The use of working memory is quite ubiquitous in human thought, but the best way to improve education using what we know about working memory is still controversial. I hope to provide some directions for research and educational practice.

What is Working Memory? An Introduction and Review

Working memory is the small amount of information that can be held in mind and used in the execution of cognitive tasks, in contrast with long-term memory, the vast amount of information saved in one’s life. Working memory is one of the most widely-used terms in psychology. It has often been connected or related to intelligence, information processing, executive function, comprehension, problem-solving, and learning, in people ranging from infancy to old age and in all sorts of animals. This concept is so omnipresent in the field that it requires careful examination both historically and in terms of definition, to establish its key characteristics and boundaries. By weaving together history, a little philosophy, and empirical work in psychology, in this opening section I hope to paint a clear picture of the concept of working memory. In subsequent sections, implications of working memory for cognitive development, learning, and education will be discussed in turn, though for these broad areas it is only feasible to touch on certain examples.

Some researchers emphasize the possibility of training working memory to improve learning and education. In this chapter, I take the complementary view that we must learn how to adjust the materials to facilitate learning and education with the working memory abilities that the learner has. Organizing knowledge, for example, reduces one’s memory load because the parts don’t have to be held in mind independently.

Take, for example, the possibility of doing some scouting ahead so that you will know what this article is about, making your task of reading easier. If you tried to read through the headings of this article, you might have trouble remembering them (placing them all in working memory) so as to anticipate how they fit together. If you read Figure 1 , though, it is an attempt to help you organize the information. If it helps you associate the ideas to one another to build a coherent framework, it should help you read by reducing the working-memory load you experience while reading. In doing so, you are building a rich structure to associate the headings with one another in long-term memory (e.g., Ericsson & Kintsch, 1995 ), which reduces the number of ideas that would have to be held independently in working memory in order to remember the organization.

Schematic diagram of the arguments in the present article.

Early History of Working Memory Research

In 1690, John Locke distinguished between contemplation, or holding an idea in mind, and memory, or the power to revive an idea after it has disappeared from the mind ( Logie, 1996 ). The holding in mind is limited to a few concepts at once and reflects what is now called working memory, as opposed to the possibly unlimited store of knowledge from a lifetime that is now called long-term memory. Working memory can be defined as the small amount of information that can be held in an especially accessible state and used in cognitive tasks.

Philosophers have long been interested in the limits of what can be contemplated, as noted by a leading British economist and logician, William Stanley Jevons. In an article in Science in 1871, he mused (p. 281): “It is well known that the mind is unable through the eye to estimate any large number of objects without counting them successively. A small number, for instance three or four, it can certainly comprehend and count by an instantaneous and apparently single act of mental attention.” Then he devised a little experiment to test this hypothesis, on himself. On each trial, he casually reached into a jar full of beans, threw several beans onto a table, and tried to estimate their number without counting. After 1,027 trials, he made no errors for sets of 3 or 4 beans, with some small errors for sets of 5 beans, and with increasing magnitudes of error as a function of set size thereafter, up to 15 beans. Despite the problematic nature of the method (in that the bean thrower was also the bean judge), the finding that normal adults typically can keep in mind only about 3 or 4 items has been replicated many times in modern research, using methods similar to Jevons (e.g., Mandler & Shebo, 1982 ) and using many other methods ( Cowan, 2001 ). The limited amount that could be held in mind at once played an important role in early experimental psychology, e.g., in the early experimental work of Hermann Ebbinghaus (1885/1913) and Wilhelm Wundt (1894/1998) . On the American front, William James (1890) wrote about a distinction between primary memory, the items in consciousness and the trailing edge of what is perceived in the world, and secondary memory, the items in storage but not currently in consciousness. Recent investigators have considered multiple possible reasons why primary memory might be limited to just a few items at once, including biological accounts based on the need to avoid confusion between concurrent objects in memory, and evolutionary and teleological accounts based on ideas about what capacity might be ideal for learning and memory retrieval ( Cowan, 2010 ; Sweller, 2011 ), but as yet the reason is unknown.

Ubiquity of the Working Memory Concept

When we say that working memory holds a small amount of information , by this term we may be referring to something as abstract as ideas that can be contemplated, or something as concrete as objects that can be counted (e.g., beans). The main point of information is that it is the choice of some things out of a greater set of possible things. One of the exciting aspects of working memory is that it may be important on so many different levels, and in so many different situations. When you are listening to language, you need to retain information about the beginning of the sentence until you can make sense of it. If you hear Jean would like to visit the third building on the left you need to recall that the actor in the sentence is Jean. Then you need to retain the verb until you know what it is she would like to visit, and you need to retain the adjective “third” until you know, third what; and all of the pieces must be put together in the right way. Without sufficient working memory, the information would be lost before you could combine it into a coherent, complete thought. As another example of how working memory is used, when doing simple arithmetic in your head, if you want to add 24 and 18 you may need to find that 4+8=12, retain the 2 while carrying the 1 over to the tens column to make 2+1+1=4 in the tens column, and integrate with the ones columns to arrive at the answer, 42. As a third example, if you are searching for your car in a parking lot, you have to remember the layout of the cars in the region you just searched so that you can avoid wasting time searching the same region again. In the jungle, a predator that turns its vision away from a scene and revisits it moments later may use working memory to detect that something in the scene has shifted; this change detection may indicate the presence of prey.

So the information in working memory can range from spoken words and printed digits to cars and future meals. It can even encompass abstract ideas. Consider whether a young child can get a good understanding of what is or is not a tiger (a matter of word category concepts, e.g., Nelson, 1974 ; Saltz, Soller, & Sigel, 1972 ). The concept is, in lay terms, a big cat with stripes. It excludes lions, which have no stripes, and it excludes zebras, which are not big cats. The child must be able to keep in mind the notion of a cat and the notion of stripes at the same time in order to grasp the tiger concept correctly. If the child thinks only of the stripes, he or she may incorrectly label a zebra as a tiger. The concept presumably starts out in working memory and, once it is learned, is transferred to long-term memory. At first, an incomplete concept might be stored in long-term memory, leading to misconceptions that are corrected later when discrepancies with further input are noticed and working memory is used to amend the concept in long-term memory. On a more abstract plane, there are more semantic issues mastered somewhat later in childhood (e.g., Clark & Garnica, 1974 ). The concept of bringing something seems to require several conditions: the person doing the bringing must have something at a location other than the speaker’s location (or future planned location), and must accompany that thing to the speaker’s location. You can ask the person to bring a salad to your house, but probably not to take a salad to your house (unless you are not there), and not to send a salad to your house (unless they are not coming along). These conditions can tax working memory. Again, the child’s initial concept transferred from working memory to long-term memory may be incomplete, and amended later when discrepancies with further input are noticed.

Working Memory: The Past 64 Years

There are several modern beginnings for the working memory concept. Hebb (1949) had an outlook on temporary memory that was more neurologically based than the earlier concept of primary memory of James (1890) . He spoke of ideas as mediated by assemblies of cells firing in a specific pattern for each idea or concept, and only a few cell assemblies would be active, with current neural firing, at any moment. This vision has played an important role in the field. An issue that is raised by this work is whether working memory should be identified with all of the active information that can be used in immediate memory tests, whether conscious or not, or whether it should be reserved to describe only the conscious information, more in the flavor of James. Given that working memory is a term usually used to explain behavioral outcomes rather than subjective reports, it is typically not restricted to conscious primary memory (e.g., see Baddeley, 1986 ; Baddeley & Hitch, 1974 ; Cowan, 1988 ). Cowan explicitly suggested that there are two aspects of working memory storage: (1) the activated portion of long-term memory, perhaps corresponding to Hebb’s active cell assemblies, and (2) within that activated portion, a smaller subset of items in the focus of attention. The activated memory would consist of a fragmented soup of all kinds of activated features (sensory, phonological, orthographic, spatial, and semantic), whereas the focus of attention would contain just a few well-integrated items or chunks.

Contributions of George Miller

Miller (1956) discussed the limitation in how many items can be held in immediate memory. In the relevant test procedure, a list of items is seen or heard and immediately afterward (that is, with no imposed retention interval), the list must be repeated verbatim. The ability to do so was said to be limited to about seven chunks, where a chunk is a meaningful unit. For example, the random digit list 582931 may have to be encoded initially as six chunks, one per digit, whereas the sequence 123654 probably can be encoded by most adults as only two chunks (an ascending triplet followed by a descending triplet). Subsequent work has suggested that the number seven is a practical result that emerges on the basis of strategies that participants use and that, when it is not possible to use chunking or covert verbal rehearsal to help performance, adults typically can retain only 3 or 4 pre-existing chunks ( Chen & Cowan, 2009 ; Cowan, 2001 ; Cowan, Rouder, Blume, & Saults, 2012 ; Luck & Vogel, 1997 ; Rouder et al., 2008 ).

The first mention I have found of the term working memory comes from a book by Miller, Galanter, and Pribram (1960) , Plans and the structure of behavior . The title itself, and the concept of organization, seems reminiscent of the earlier work by Hebb (1949) , The organization of behavior . Miller et al. observed that daily functioning in the world requires a hierarchy of plans. For example, your plan to do well at work requires a sub-plan to be there at time in the morning, which in turn may require sub-plans to eat breakfast, shower, get dressed, gather work materials, and so on. Each of these plans also may have sub-plans, and you may have competing plans (such as choosing an after-work activity, calling your mother, or acquiring food for dinner). Our working memory was said to be the mental faculty whereby we remember the plans and sub-plans. We cannot think about all of them at once but we might, for example, keep in mind that the frying pan is hot while retrieving a knife from the drawer, and we may keep bringing to mind the approximate time so as not to be late. Working memory was said to be the facility that is used to carry out one sub-plan while keeping in mind the necessary related sub-plans and the master plan.

Contributions of Donald Broadbent

In Great Britain, Broadbent’s (1958) book helped to bring the conversation out of the behaviorist era and into an era of cognitive psychology. In a footnote within the book, he sketched a rough information processing diagram that showed information progressing from a sensory type of store that holds a lot of information briefly, through an attention filter to essentially a working memory that holds only a few items, to a long-term memory that is our storehouse of knowledge accumulated through a lifetime. The empirical basis for the model came largely from his work with selective attention, including many dichotic listening studies in which the task was to listen to the message from one ear and ignore the message from the other ear, or report both messages in some order. The motivation for this kind of research came largely from practical issues provoked by World War II, such as how to help a pilot listen to his own air traffic control message while ignoring messages meant for other pilots but presented in the same channel. An important theoretical outcome, however, was the discovery of a difference between a large-capacity but short-lived sensory memory that was formed regardless of attention, and a longer-lived but small-capacity abstract working memory that required attention.

Contributions of Alan Baddeley and Graham Hitch

Miller et al. (1960) may have devised the term working memory but they were not the predominant instigator of the work that has occurred subsequently in the field. Google Scholar does show it with over 5,600 citations. A chapter by Baddeley and Hitch (1974) , though, is listed with over 7,400 citations and a 1992 Science article summarizing that approach has over 14,500 citations. In the 1974 chapter, the term working memory was used to indicate a system of temporary memory that is multifaceted, unlike the single store such as James’ primary memory, or the corresponding box in Broadbent’s (1958) model, or an elaborated version of it as in the model of Atkinson and Shiffrin (1968) model, none of which would do. In fact, a lot of investigators in the 1960’s proposed variations of information processing models that included a single short-term memory store, and Baddeley often has referred to these together, humorously, as the “modal model,” providing a sketch of it with sensory memory, short-term memory, and long-term memory boxes as in the Broadbent and the Atkinson/Shiffrin models. (When the humor and the origin of the phrase “modal model” are forgotten, yet the phrase is still widely used, it seems sad somehow.)

The main point emphasized by Baddeley and Hitch (1974) is that there were diverse effects that appeared to implicate short-term memory, but that did not converge to a single component. Phonological processing interfered most with phonological storage, visual-spatial processing interfered with visual-spatial storage, and a working memory load did not seem to interfere much with superior memory for the end of a list, or recency effect. Conceptual learning did not depend heavily on the type of memory that was susceptible to phonological similarity effects, and a patient with a very low memory span was still able to learn new facts. To account for all of the dissociations, they ended up concluding that there was an attention-related control system and various storage systems. These included a phonological system that also included a covert verbal rehearsal process, and a visual-spatial storage system that might have its own type of non-verbal rehearsal. In the 1974 version of the theory, there were attention limits on the storage of information as well as on processing. In a 1986 book, Baddeley eliminated the attention-dependent storage but in a 2000 paper, a new component was added in the form of an episodic buffer. This buffer might or might not be attention-dependent and is responsible for holding semantic information for the short term, as well as the specific binding or association between phonological and visual-spatial information. Baddeley and Hitch called the assembly or system of storage and processing in service of holding information in an accessible form working memory, the memory one uses in carrying out cognitive tasks of various kinds (i.e., cognitive work).

Model of Cowan (1988)

Through the years, there were several other proposals that alter the flavor of the working memory proposal. Cowan (1988) was concerned with how we represent what we know and do not know about information processing. The “modal models” of which Baddeley has spoken began with Broadbent’s (1958) model in which the boxes were shown to be accessed in sequence, comparable to a computer flow chart: first sensory memory, then an attention filter, then short-term memory, and then long-term memory. Atkinson & Shiffrin (1968) preserved the flow chart structure but added recursive entry into the boxes, in the form of the control processes. Baddeley and Hitch (1974) and Baddeley (1986) instead used a processing diagram in which the boxes could be accessed in parallel. One presumably could enter some information into phonological storage while concurrently entering other information into visual-spatial storage, with interacting modules and concurrent executive control.

Cowan (1988 , 1995 , 1999 , 2001 , 2005 ) recoiled a bit from the modules and separate boxes, partly because they might well form an arbitrarily incomplete taxonomy of the systems in the brain. (Where would spatial information about sound go? Where would touch information go? These types of unanswered questions also may have helped motivate the episodic buffer of Baddeley, 2000 .) There could be multiple modules, but because we do not know the taxonomy, they were all thrown into the soup of activated long-term memory. Instead of separate boxes, I attempted to model on a higher level at which distinctions that were incomplete were not explicitly drawn into the model, and mechanisms could be embedded in other mechanisms. Thus, there was said to be a long-term memory, a subset of which was in an activated state (cf. Hebb, 1949 ), and within that, a smaller subset of which was in the focus of attention (cf. James, 1890 ). Dissociations could still occur on the basis of similarity of features; two items with phonological features will interfere with one another, for example, more than one item with phonological features and another item with only visual-spatial features. The model still included central executive processes.

Compared to Baddeley and Hitch (1974) , Cowan (1988) also placed more emphasis on sensory memory. It is true that printed letters, like spoken letters, are encoded with speech-based, phonological features that can be confused with each other in working memory (e.g., Conrad, 1964 ). Nevertheless, there is abundant other evidence that lists presented in a spoken form are remembered much better, in particular at the end of the list, than verbal lists presented in printed form (e.g., Murdock & Walker, 1969 ; Penney, 1989 ).

The attention filter also was internalized in the model of Cowan (1988) . Instead of information having to pass through a filter, it was assumed that all information activates long-term memory to some degree. The mind forms a neural model of what it has processed. This will include sensory information for all stimuli but, in the focus of attention, much more semantic information than one finds for unattended information. Incoming information that matches the current neural model becomes habituated, but changes that are perceived cause dishabituation in the form of attentional orienting responses toward the dishabituated stimuli (cf. Sokolov, 1963 ). Such a system has properties similar to the attenuated filtering model of Treisman (1960) or the pertinence model of Normal (1968) . Attention is controlled in this view dually, often with a struggle between voluntary executive control and involuntary orienting responses.

How consistent is Cowan (1988) with the Baddeley and Hitch model? Contributions of Robert Logie

With the addition of the episodic buffer, the model of Baddeley and Hitch makes predictions that are often similar to those of Cowan (1988) . There still may be important differences, though. An open question is whether the activated portion of long-term memory of Cowan (1988) functionally serves the same purpose as the phonological and visual-spatial buffers of Baddeley and Hitch (1974) and Baddeley (1986) . Robert Logie and colleagues argue that this cannot be, inasmuch as visual imagery and visual short-term memory are dissociated ( Borst, Niven, & Logie, 2012 ; Logie & van der Meulen, 2009 ; van der Meulen, Logie, & Della Sala, 2009 ). Irrelevant visual materials interfere with the formation of visual imagery but not with visual storage, whereas tapping in a spatial pattern interferes with visual storage but not the formation of visual images. According to the model that these sources put forward, visual imagery involves activation of long-term memory representations, whereas visual short-term storage is a separate buffer. Although this is a possibility that warrants further research, I am not yet convinced. There could be other reasons for the dissociation. For example, in the study of van der Meulen et al., the visual imagery task involved detecting qualities of the letters presented (curved line or not, enclosed space or not, etc.) and these qualities could overlap more with the picture interference; whereas the visual memory task involved remembering letters in upper and lower case visually, in the correct serial order, and the serial order property may suffer more interference from tapping in a sequential spatial pattern. Testing of the generality of the effects across tasks with different features is needed.

Other models of cross-domain generality

One difference between the Baddeley (1986) framework and that of Cowan (1988) was that Cowan placed more emphasis on the possibility of interference between domains. There has been a continuing controversy about the extent to which verbal and nonverbal codes in working memory interfere with one another (e.g., Cocchini, Logie, Della Sala, MacPherson, & Baddeley, 2002 ; Cowan & Morey, 2007 ; Fougnie & Marois, 2011 ; Morey & Bieler, 2013 ). The domain-general view has extended to other types of research. Daneman and Carpenter (1980) showed that reading and remembering words are tasks that interfere with one another, with the success of remembering in the presence of reading a strong correlate of reading comprehension ability. Engle and colleagues (e.g., Engle, Tuholski, Laughlin, & Conway, 1999 ; Kane et al., 2004 ) showed that this sort of effect does not just occur with verbal materials, but occurs even with storage and processing in separate domains, such as spatial recall with verbal memory. They attributed individual differences primarily to the processing tasks and the need to hold in mind task instructions and goals while suppressing irrelevant distractions.

Barrouillet and colleagues (e.g., Barrouillet, Portrat, & Camos, 2011 ; Vergauwe, Barrouillet, & Camos, 2010 ) emphasized that the process of using attention to refresh items, no matter whether verbal or nonverbal in nature, takes time and counteracts decay. They provided complex tasks involving concurrent storage and processing, like Daneman and Carpenter and like Engle and colleagues. The key measure is cognitive load, the proportion of time that is taken up by the processing task rather than being free for the participant to use to refresh the representations of items to be remembered. The finding of Barrouillet and colleagues has been that the effect of cognitive load on the length of list that can be recalled, or memory span, is a negative linear (i.e., deleterious) effect. They do also admit that there is a verbal rehearsal process that is separate from attentional refreshing, with the option of using either mode of memory maintenance depending on the task demands ( Camos, Mora, & Oberauer, 2011 ), but there is more emphasis on attentional refreshing than in the case of Baddeley and colleagues, and the approach therefore seems more in keeping with Cowan (1988) with its focus of attention (regarding refreshing see also Cowan, 1992 ).

Ongoing controversies about the nature of working-memory memory limits

There are theoretically two basic ways in which working memory could be more limited than long-term memory. First, It could be limited in terms of how many items can be held at once, a capacity limit that Cowan (1998, 2001 ) tentatively ascribes to the focus of attention. Second, it could be limited in the amount of time for which an item remains in working memory when it is no longer rehearsed or refreshed, a decay limit that Cowan (1988) ascribed to the activated portion of long-term memory, the practical limit being up to about 30 seconds depending on the task.

Both of these limits are currently controversial. Regarding the capacity limit, there is not much argument that, within a particular type of stimulus coding (phonological, visual-spatial, etc.), normal adults are limited to about 3 or 4 meaningful units or chunks. The debate is whether the limit occurs in the focus of attention, or because materials of similar sorts interfere with one another (e.g., Oberauer, Lewandowsky, Farrell, Jarrold, & Greaves, 2012 ). In my recent, still-unpublished work, I suggest that the focus of attention is limited to several chunks of information, but that these chunks can be off-loaded to long-term memory and held there, with the help of some attentional refreshing, while the focus of attention is primarily used to encode additional information.

Regarding the memory loss or decay limit, some studies have shown no loss of information for lists of printed verbal materials across periods in which rehearsal and refreshing have apparently been prevented ( Lewandowsky, Duncan, & Brown, 2004 ; Oberauer & Lewandowsky, 2008 ). Nevertheless, for arrays of unfamiliar characters followed by a mask to eliminate sensory memory, Ricker and Cowan (2010) did find memory loss or decay (cf. Zhang & Luck, 2009 ). In further work, Ricker et al. (in press) suggested that the amount of decay depends on how well the information is consolidated in working memory (cf. Jolicoeur & Dell'Acqua, 1998 ). Given that the time available for refreshing appeared to be inversely related to the cognitive load, the consolidation process that seems critical is not interrupted by a mask but continues after it. This consolidation process could be some sort of strengthening of the episodic memory trace based on attentional refreshing in the spirit of Barrouillet et al. (2011) . If so, the most important effect of this refreshing would not be to reverse the effects of decay temporarily, as Barrouillet et al. proposed, but rather to alter the rate of decay itself. Our plans for future research include investigation of these possibilities.

Long-term working memory

It is clear that people function quite well in complex environments in which detailed knowledge must be used in an expert manner, despite a severe limit in working memory to a few ideas or items at once. What is critical in understanding this paradox of human performance is that each slot in working memory can be filled with a concept of great complexity, provided that the individual has the necessary knowledge in long-term memory. This point was made by Miller (1956) in his concept of combining items to form larger chunks of information, with the limit in working memory found in the number of chunks, not the number of separate items presented for memorization. Ericsson and Kintsch (1995) took this concept further by expanding the definition of working memory to include relevant information in long-term memory.

Although we might quibble about the best definition of working memory, it seems undeniable that long-term memory is often used as Ericsson and Kintsch (1995) suggest. An example is what happens when one is holding a conversation with a visitor that is interrupted by a telephone call. During the call, the personal conversation with one’s visitor is typically out of conscious working memory. After the call, however, with the visitor serving as a vivid cue, it is often possible to retrieve a memory of the conversation as a recent episode and to remember where this conversation left off. That might not be possible some days later. This use of long-term memory to serve a function similar to the traditional working memory, thus expanding the person’s capabilities, was termed long-term working memory by Ericsson and Kintsch. Cowan (1995) alluded to a similar use of long-term memory for this purpose but, not wanting to expand the definition of working memory, called the function virtual short-term memory, meaning a use of long-term memory in a way that short-term memory is usually used. It is much like the use of computer memory that allows the computer to be turned off in hibernation mode and later returned to its former state when the memory is retrieved.

Given the ability of humans to use long-term memory so adeptly, one could ask why we care about the severe working memory capacity limit at all. The answer is that it is critical when there is limited long-term knowledge of the topic. In such circumstances, the capacity of working memory can determine how many items can be held in mind at once in order to use the items together, or to link them to form a new concept in long-term memory. This is the case in many situations that are important for learning and comprehension. One simple example of using items together is following a set of instructions, e.g., to a preschool child, put your drawing in your cubby and then go sit in the circle . Part of that instruction may be forgotten before it is carried out and teachers must be sensitive to that possibility. A simple example of linking items together is in reading a novel, when one listens to a description of a character and melds the parts of the description to arrive at an overall personality sketch that can be formed in long-term memory. Inadequate use of working memory during reading may lead to the sketch being incomplete, as some descriptive traits are inadvertently ignored. Knowledge of this working memory limit can be used to improve one’s writing by making it easier to remember and comprehend.

Paas and Sweller (2012) bring up the distinction between biologically primary and secondary knowledge ( Geary, 2008 ) and suggest (p. 29) that “Humans are easily able to acquire huge amounts of biologically primary knowledge outside of educational contexts and without a discernible working memory load.” Examples they offered were the learning of faces and learning to speak. It may well be the case that individual faces or spoken words quickly become integrated chunks in long-term memory (and, I would add, the same seems true for objects in domains of learned expertise, e.g., written words in adults). Nevertheless, the biologically-primary components are used in many situations in which severe capacity limits do apply. In these situations, the added memory demand is considered biologically secondary. An example is learning which face should be associated with which name. If four novel faces are shown on a screen and their names are vocally presented, these name-face pairs cannot be held in working memory at once, so it is difficult to retain the information and it often takes additional study of one pair at a time to remember the name-face pairing.

Specific mathematical models

Here I have been selective in examining models of working memory that are rather overarching and verbally specified. By limiting the domain of applicability and adding some processing assumptions, other researchers throughout the years have been able to formulate models that make mathematical predictions of performance in specific situations. We have learned a lot from them but they are essentially outside of the scope of this review given limited space and given my own limitations. For examples of such models see Brown, Neath, & Chater, 2007 ; Burgess & Hitch, 1999 ; Cowan et al., 2012 ; Farrell & Lewandowsky, 2002 ; Hensen, 1998 ; Murdock, 1982; Oberauer & Lewandowsky, 2011 ). The importance of these models is that they make clear the consequences of our theoretical assumptions. In order to make quantitative predictions, each mathematical assumption must be made explicit. It is sometimes found that the effects of certain proposed mechanisms, taken together, are not what one might assume from a purely verbal theory. Of course, some of the assumptions that one must make to eke out quantitative predictions may be unsupported, so I believe that the best way forward in the field is to use general verbal, propositional thinking some of the time and specific quantitative modeling other times, working toward a convergence of these methods toward a common theory.

Summary: Status of Working Memory

The progress in this field might be likened to an upward spiral. We make steady progress but meanwhile, we go in circles. The issues of the nature of working memory limits have not changed much from the early days. Why is the number of items limited? Why is the duration limited? What makes us forget? How is it related to the conscious mind and to neural processes? These questions are still not answered. At the same time, we have agreement about what can be found in particular circumstances. Set up the stimuli one way, and there is interference between modalities. Set it up another way and there appears to be much less interference. Set it up one way and items are lost rapidly across time. Set it up a different way, and there is much less loss. There are brain areas associated with the focus of attention and with working memory across modalities ( Cowan, 2011 ; Cowan, Li et al., 2011 ; Todd & Marois, 2004 ; Xu & Chun, 2006 ). This is progress awaiting an adequate unifying theory.

What we do know has practical implications. To avoid overtaxing an individual’s working-memory capabilities, one should avoid presenting more than a few items or ideas at once, unless the items can be rapidly integrated. One should also avoid making people hold on to unintegrated information for a very long time. For example, I could write a taxing sentence like, It is said that, if your work is not overwhelming, your car is in good repair, and the leaves have changed color, it is a good time for a fall vacation . However, that sentence requires a lot from the reader’s working memory. I could reduce the working memory load by not making you wait for the information that provides the unifying theme, keeping the working memory load low: It is said that a good time for a fall vacation is when your work is not overwhelming, your car is in good repair, and the leaves have changed color .

Working Memory and Cognitive Development

There is no question that working-memory capabilities increase across the life span of the individual. In early tests of maturation (e.g., Bolton, 1892 ), and to this day in tests of intelligence, children have been asked to repeat lists of random digits. The length of list that can be successfully repeated on some predefined proportion of trials is the digit span. It increases steadily with childhood maturation, until late childhood. When the complexity of the task is increased, the time to adult-like performance is extended a bit further, with steady improvement throughout childhood (for an example see Gathercole, Pickering, Ambridge, & Wearing, 2004 ).

As we saw in the introductory section, clear practical findings do not typically come with a clear understanding of the theoretical explanation. There have been many explanations over the years for the finding of increasing memory span with age (e.g., see Bauer & Fivush, in press ; Courage & Cowan, 2009 ; Kail, 1990 ). These explanations may lead to differing opinions of the best course for learning and education, as well.

Explanations Based on Capacity

Explanations of intellectual growth based on working memory capacity stem from what has been called the neoPiagetian school of thought. Jean Piaget outlined a series of developmental stages, but with no known underlying reason for the progression between stages. Pascual-Leone and Smith (1969) attributed the developmental increases to increases in the number of items that could be held in mind at once.

The theory becomes more explicit with the contributions of Halford, Phillips, and Wilson (1998) and Andrews and Halford (2002) . They suggests that it is the number of associations between elements that is restricted and that this matters because it limits the complexity of thought. In my example above, the concept of a tiger versus lion versus zebra requires concurrent consideration of the animal’s shape and presence or absence of stripes. Similarly, addition requires the association between three elements: the two elements being added and the sum. A concept like bigger than is a logical relation requiring three slots, e.g., bigger than (dog, elephant) . Ratios require the coordination of four elements (e.g., 4/6 is equivalent to 6/9) and therefore are considerably harder to grasp, according to the theory (see Halford, Cowan, & Andrews, 2007 ).

This concept is quite promising and might even appear to be “the only game in town” when it comes to trying to understand the age limits on children’s ability to comprehend ideas of various levels of complexity. One problem with it is that it is not always straightforward to determine the arity of a concept, or number of ideas that must be associated. For example, a young child might understand the concept big(elephant) and then might be able to infer that elephants are bigger than dogs, without being able to use the concept of bigger than in a consistent manner more generally. The concept from Miller (1956) that items can be combined using knowledge to form larger chunks also applies to associations, and it is not clear how to be sure that the level of complexity actually is what it is supposed to be. Knowledge allows some problems to be solved with less working memory requirement.

Explanations Based on Knowledge

It is beyond question that knowledge increases with age. Perhaps this knowledge increase is the sole reason for developmental change in working memory, it has been argued. Chi (1978) showed that children with an expertise in the game of chess could remember chess configurations better than adults with no such expertise. The expert children presumably could form larger chunks of chess pieces, greatly reducing the memory load. Case, Kurland, and Goldberg (1982) gave adults materials that were unfamiliar and found that both the speed of item identification and the memory span for those materials closely resembled what was found for 6-year-olds on familiar materials. The implication was that the familiarity with the materials determines the processing speed, which in turn determines the span.

Explanations Based on Processing Speed and Strategies

Case et al. (1982) talked of a familiarity difference leading to a speed difference. Others have suggested that, more generally, speed of processing increases with age in childhood and decrease again with old age (e.g., Kail & Salthouse, 1994 ). This has led to accounts of working memory improvement based on an increased rate of covert verbal rehearsal ( Hulme & Tordoff, 1989 ) or increased rate of attentional refreshing ( Barrouillet, Gavens, Vergauwe, Gaillard, & Camos, 2009 ; Camos & Barrouillet, 2011 ). At the lower end of childhood, it has been suggested on the basis of various evidence that young children do not rehearse at all ( Flavell, Beach, & Chinsky, 1966 ; Garrity, 1975 ; Henry, 1991 ) or do not rehearse in a sufficiently sophisticated manner that is needed to assist in recall ( Ornstein & Naus, 1978 ). When rehearsal aloud is required, the result suggest that the most recently rehearsed items are recalled best ( Tan & Ward, 2000 ).

This view that rehearsal is actually important has been opposed recently. It is not clear that rehearsal must be invoked to explain performance ( Jarrold & Citroën, 2013 ) and if rehearsal takes place, it is not clear exactly what the internal processes are (e.g., cumulative repetition of the list? Repetition of each item as it is presented?).

In the case of using attention to refresh information, an interesting case can be made. Children who are too young (about 4 years of age and younger) do not seem to use attention to refresh items. For them, the limit in performance depends on the duration of the retention interval. For older children and adults, who are able to refresh, it is not the absolute duration but the cognitive load that determines performance ( Barrouillet et al., 2011 ). The “phase change” in performance that is observed here with the advent of refreshing is perhaps comparable to the phase change that is seen with the advent of verbal rehearsal ( Henry, 1991 ), though the evidence may be stronger in the case of refreshing.

Re-assessment of Capacity Accounts

We have seen that there are multiple ways in which children’s working memory performance gets better with maturity. There are reasons to care about whether the growth of capacity is primary, or whether it is derived from some other type of development. For example, if the growth of capacity results only from the growth of knowledge, then it should be possible to teach any concept at any age, if the concept can be made familiar enough. If capacity differences come from speed differences, it might be possible to allow more time by making sure that the parts to be incorporated into a new concept are presented sufficiently slowly.

We have done a number of experiments suggesting that there is something to capacity that changes independent of these other factors. Regarding knowledge, relevant evidence was provided by Cowan, Nugent, Elliott, Ponomarev, and Saults (1999) in their test of memory for digits that were unattended while a silent picture-rhyming game was carried out. The digits were attended only occasionally, when a recall cue was presented about 1 s after the last digit. The performance increase with age throughout the elementary school years was just as big for small digits (1, 2, 3), which are likely to be familiar, as for large digits (7, 8, 9), which are less familiar. Gilchrist, Cowan, and Naveh-Benjamin (2009) further examined memory for lists of unrelated, spoken sentences in order to distinguish between a measure of capacity and a measure of linguistic knowledge. The measure of capacity was an access rate, the number of sentences that were at least partly recalled. The measure of linguistic knowledge was a completion rate, the proportion of a sentence that was recalled, provided that at least part of it was recalled. This sentence completion rate was about 80% for both first and sixth grader children, suggesting that for these simple sentences, there was no age difference in knowledge. Nevertheless, the number of sentences accessed was considerably smaller in first-grade children than in sixth-grade children (about 2.5 sentences vs. 4 sentences). I conclude, tentatively at least, that knowledge differences cannot account for the age difference in working memory capacity.

We have used a different procedure to help rule out a number of factors that potentially could underlie the age difference in observed capacity. It is based on a procedure that has been well-researched in adults ( Luck & Vogel, 1997 ). On each trial of this procedure, an array of simple items (such as colored squares) is presented briefly and followed by a retention interval of about 1 s, and then a single probe item is presented. The latter is to be judged identical to the array item from the same location, or a new item. This task is convenient partly because there are mathematical ways to estimate the number of items in working memory ( Cowan, 2001 ). If k items are in working memory and there are N items in the array, the likelihood that the probed item is known is k/N , and a correct response can also come from guessing. It is possible to calculate k , which for this procedure is equal to N ( h-f ), where h refers the proportion of change trials in which the change was detected (hits) and f refers to the proportion of no-change trials in which a change was incorrectly reported (false alarms).

One possibility is that younger children remember less of the requested information because they attend to more irrelevant information, cluttering working memory (for adults, cf. Vogel, McCollough, & Machizawa, 2005 ). To examine this, Cowan, Morey, AuBuchon, Zwilling, and Gilchrist (2010) presented both colored circles and colored triangles and instructed participants to pay closer attention to one shape, which was tested on 80% of the trials in critical blocks. When there were 2 triangles and 2 circles, memory for the more heavily-attended shape was better than memory for the less-attended shape, to the same extent in children in Grades 1–2 and Grades 6–7, and in college students. Yet, the number of items in working memory was much lower in children in Grades 1–2 than in the two older groups. It did not seem that the inability to filter out irrelevant information accounted for the age difference in capacity.

Another possibility is that in Cowan et al. (2010) , the array items occurred too fast for the younger children to encode correctly. To examine this, Cowan, AuBuchon, Gilchrist, Ricker, & Saults (2011) presented the items one at a time at relatively slow, a 1-item-per-second rate. The results remained the same as before. In some conditions, the participant had to repeat each color as it was presented or else say “wait” to suppress rehearsal; this articulatory manipulation, too, left the developmental effect unchanged. It appears that neither encoding speed nor articulation could account for the age differences. So we believe that age differences in capacity may be primary rather than derived from another process.

Age differences in capacity still could occur because of age differences in the speed of a rapid process of refreshment, and from the absence of refreshment in young children ( Camos & Barrouillet, 2011 ). Alternatively, it could occur because of age differences in some other type of speed, neural space, or efficiency. This remains to be seen but at least we believe that there is a true maturational change in working memory capacity underlying age differences in the ability to comprehend materials of different complexity. This is in addition to profound effects of knowledge acquisition and the ability to use strategies.

The use of strategies themselves may be secondary to the available working memory resources to carry out those strategies. According to the neoPiagetian view of Pascual-Leone and Smith (1969) , for example, the tasks themselves share resources with the data being stored. Cowan et al. (2010) found that when the size of the array to be remembered was large (3 more-relevant and 3 less-relevant items, rather than 2 of each) then young children were no longer able to allocate more attention to the more-relevant items. The attentional resource allocated to the items in the array was apparently deducted from the resource available to allocate attention optimally.

In practical terms, it is worth remembering that several aspects of working memory are likely to develop: capacity, speed, knowledge, and the use of strategies. Although it is not always easy to know which process is primary, these aspects of development all should contribute in some way to our policies regarding learning and education.

Working Memory and Learning

In early theories of information processing, up through the current period, working memory was viewed as a portal to long-term memory. In order for information to enter long-term memory in a form that allows later retrieval, it first must be present in working memory in a suitable form. Sometimes that form appears modality-specific. For example, Baddeley, Papagno, and Vallar (1988) wondered how it could be that a patient with a very small verbal short-term memory span, 2 or 3 digits at most, could function so well in most ways and exhibit normal learning capabilities. The answer turned out to be that she displayed a very selective deficit: she was absolutely unable to learn new vocabulary. This finding led to a series of developmental studies showing that individual differences in phonological memory are quite important for differences in word-learning capability in both children and adults ( Baddeley, Gathercole, & Papagno, 1998 ; Gathercole & Baddeley, 1989 , 1990 ).

Aside from this specific domain, there are several ways in which working memory can influence learning. It is important to have sufficient working memory for concept formation. The control processes and mnemonic strategies used with working memory also are critical to learning.

Working Memory and Concept Formation

Learning might be thought of in an educational context as the formation of new concepts. These new concepts occur when existing concepts are joined or bound together. Some of this binding is mundane. If an individual knows what the year 1776 means and also what the Declaration of Independence is (at least in enough detail to remember the title of the declaration), then it is possible to learn the new concept that the Declaration of Independence was written in the year 1776. Other times, the binding of concepts may be more interesting and there may be a new conceptual leap involved. For example, a striped cat is a tiger. As another simple example, to understand what a parallelogram is, the child has to understand what the word parallel means, and further to grasp that two sets of parallel lines intersect with one another. The ideas presumably must co-exist in working memory for the concept to be formed.

For the various types of concept formation, then, the cauldron is assumed to be working memory. According to my own view, the binding of ideas occurs more specifically in the focus of attention. We have taken a first step toward verifying that hypothesis. Cowan, Donnell, and Saults (in press) presented lists of words with an incidental task: to report the most interesting word in each presented list. Later, participants completed a surprise test in which they were asked whether pairs of words came from the same list; the words were always one or two serial positions apart in their respective lists, but sometimes were from the same list and sometimes from different lists. The notion was that the link between the words in the same list would be formed only if the words had been in the focus of attention at the same time, which was much more likely for short lists than for long lists. In keeping with this hypothesis, performance was better for words from short lists of 3 items (about 59%) than for words from lists of 6 or 9 items (about 53%). This is a small effect, but it is still important that there was unintentional learning of the association between items that were together in the focus of attention just once, when there was no intention of learning the association.

The theory of Halford et al. (1998) may be the best articulated theory suggesting why a good working memory is important for learning. (In this discussion, a “good” working memory is simply one that can keep in mind sufficient items and their relations to one another to solve the problem at hand, which may require a sufficient combination of capacity, speed, knowledge, and available strategies.) More complex concepts require that one consider the relationship between more parts. A person’s working memory can be insufficient for a complex concept. It may be possible to memorize that concept with less working memory, but not truly to understand the concept and work with it. Take, for example, use of the concept of transitivity in algebra. If a+b=c+d and c+d=e , then we can conclude that a+b=e because equality is transitive. Yet, a person who understands the rules of algebra still would not be able to draw the correct inference if that person could not concurrently remember the two equations. Even if the equations are side by side on the page, that does not mean that they necessarily can be encoded into working memory at the same time, which is necessary in order to draw the inference. Lining up the equations vertically for the learner and then inviting the learner to apply the rule by rote is a method that can be used to reduce the working memory load, perhaps allowing the problem to be solved. However, working out the problem that way will not necessarily produce the insight needed to set up a new problem and solve it, because setting up the problem correctly requires the use of working memory to understand what should be lined up with what. So if the individual does not have sufficient working memory capacity, a rote method of solution may be helpful for the time being. More importantly, though, the problem could be set up in a more challenging manner so that the learner is in the position of having to use his or her working memory to store the information. By doing so, the hope is that successful solution of the problem then will result in more insight that allows the application of the principles to other problems. That, in fact, is an expression of the issues that may lead to the use of word problems in mathematics education.

Working Memory and Control Processes

Researchers appear to be in fairly good agreement that one of the most important aspects of learning is staying on task. If one does not stick to the relevant goals, one will learn something perhaps, but it will not be the desired learning. Individuals who test well on working memory tasks involving a combination of storage and processing have been shown to do a better job staying on task.

A good experimental example of how staying on task is tied to working memory is one carried out by Kane and Engle (2003) using a well-known task designed long ago by John Ridley Stroop. In the key condition within this task, one is to name the color of ink in which color words are written. Sometimes, the color of ink does not match the written color and there is a tendency to want to read the word instead of naming the color. This effect can be made more treacherous by presenting stimuli in which the word and color match on most trials, so that the participant may well lapse into reading and lose track of the correct task goal (naming the color of ink). What that happens, the result is an error or long delay on the occasional trials for which the word and ink do not match. Under those circumstances, the individuals who are more affected by the Stroop conditions are those with relatively low performance on the operation span test of working memory (carrying out arithmetic problems while remembering words interleaved with those problems).

In more recent work, Kane et al. (2007) has shown that low-span individuals have more problems attending in daily life. Participants carried devices that allowed them to respond at unpredictable times during the day, reporting what they were doing, what they wanted to be doing, and so on. It was found that low-span individuals were more likely to report that their minds were wandering away from the tasks on which they were trying to focus attention. This, however, did not occur on all tasks. The span-related difference in attending was only for tasks in which they reported that they wanted to pay attention. When participants reported that they were bored and did not want to pay attention, mind-wandering was just as prevalent for high spans as for low spans.

Although this work was done on adults, it has implications for children as well. Gathercole, Lamont, and Alloway (2006) suggest that working memory failures appear to be a large part of learning disabilities. Children who were often accused of not trying to follow directions tested out as children with low working memory ability. They were often either not able to remember instructions or not able to muster the resources to stick to the task goal and pay attention continually, for the duration needed. Children with various kinds of learning and language disability generally test below grade level on working memory procedures, and children with low working memory and executive function don’t do well in school (e.g., Sabol & Pianta, 2012 ).

Of course, central executive processes must do more than just maintain the task goal. The way in which information is converted from one form to another, the vigilance with which the individual searches for meaningful connections between elements and new solutions, and self-knowledge about what areas are strong or weak all probably play important roles in learning.

Working Memory and Mnemonic Strategies

There also are special strategies that are needed for learning. For example, a sophisticated rehearsal strategy for free recall of a list involves a rehearsal method that is cumulative. If the first word on the list is a cow, the second is a fish, and the third a stone, one ideally should rehearse cumulatively: cow…cow, fish….cow, fish, stone … and so on ( Ornstein & Naus, 1978 ). Cowan, Saults, Winterowd, and Sherk (1991) showed that young children did not carry out cumulative rehearsal the way older children do and could not easily be trained to do so, but that their memory improved when cumulative rehearsal was overtly supported by cumulative presentation of stimuli.

For long-term learning, maintenance rehearsal is not nearly as effective a strategy as elaborative rehearsal, in which a coherent story is made on the basis of the items; this takes time but results in richer associations between items, enhancing long-term memory provided that there is time for it to be accomplished (e.g., Craik & Watkins, 1973 ).

In addition to verbal and elaborative rehearsal, Barrouillet and colleagues (2011) have discussed attentional refreshing as a working-memory maintenance process. We do not yet know what refreshing looks like on a moment-to-moment basis or what implications this kind of maintenance strategy has for long-term learning. It is a rich area for future research.

The most general mnemonic strategy is probably chunking ( Miller, 1956 ), the formation of new associations or recognition of existing ones in order to reduce the number of independent items to keep track of in working memory. The power of chunking is seen in special cases in which individuals have learned to go way beyond the normal performance. Ericsson, Chase, and Faloon (1980) studied an individual who learned, over the course of a year, to repeat lists of about 80 digits from memory. He learned to do so starting with a myriad of athletic records that he knew so that, for example, 396 might be recoded as a single unit, 3.96 minutes, a fairly fast time for running the mile. After applying this intensive chunking strategy in practice for a year, a list of 80 digits could be reduced to several sub-lists, each with associated sub-parts. The idea would be that the basic capacity has not changed but each working-memory slot is filled with quite a complex chunk. In support of this explanation, individuals of this sort still remain at base level (about 7 items) for lists of items that were not practiced in this way, e.g., letters. (For a conceptual replication see Ericsson, Delaney, Weaver, & Mahadevan, 2004 ; Wilding, 2001 )

Although we cannot all reach such great heights of expert performance, we can do amazing things using expertise. For example, memorization of a song or poem is not like memorization of a random list of digits because there are logical connections between the words and between the lines. A little working memory then can go a long way.

The importance of a good working memory comes in when something new is learned, and logical connections are not yet formed so the working memory load is high. When there are not yet sufficient associations between the elements of a body of material, working memory is taxed until the material can be logically organized into a coherent structure. Working memory is thought to correlate most closely with fluid intelligence, the type of intelligence that involves figuring out solutions to new problems (e.g., Wilhelm & Engle, 2005 ). However, crystallized intelligence, the type of intelligence that involves what you know, also is closely related to fluid intelligence. The path I suggest here is that a good working memory assists in problem-solving (hence fluid intelligence); fluid intelligence and working memory then assist in new learning (hence crystallized intelligence).

Working Memory and Education

We have sketched the potential relation between working memory and learning. How is that to be translated into lessons for education? There is a large and diverse literature on this topic. As a starting point to illustrate this diversity, I will describe the chapters chosen for the book, Working memory and education ( Pickering, 2006 ). After an introductory chapter on working memory (A. Baddeley), the book includes two chapters on the relation between working memory and reading (one by P. de Jong and another by K. Cain). There is a chapter on the relation between working memory and mathematics education (R. Bull and K.A. Espy), learning disabilities (H.L. Swanson), attention disorders (K. Cornish and colleagues), and deafness (M. Keehner & J. Atkinson). Other chapters cover more general topics, including the role of working memory in the classroom (S. Gathercole and colleagues), the way to assess working memory in children (S. Pickering), and sources of working memory deficit (M. Minear and P. Shah). It is clear that many avenues of research relate working memory to education, and I cannot travel along all of them in this review.

To organize a diverse field, what I can do is to distinguish between several different basic approaches have been tried. First, one can try to teach to the level of the learner’s working memory. The points described in the article up to this point should be kept in mind when one is trying to discern and understand what a particular learner can and cannot do. Second, one can try to use training exercises to improve working memory, which, investigators have hoped, would allow a person to be able to learn more and solve problems more successfully. The message I would give here is to be wary, given the rudimentary state of the evidence in a difficult field and the plethora of companies selling working memory training exercises. Third, one might contemplate the role of working memory for the most critical goals of education, in a broad sense. These topics will be examined one at a time.

Teaching to the Level of Working Memory

The classic adaptation of education to cognitive development and the needs of learning has been to try to adjust the materials to fit the learner. For example, there has been considerable discussion of the need to delay teaching concepts of arithmetic at least until the children understand the basic underlying concept of one-to-one correspondence; that is, the idea that there are different numbers in a series and that each number is assigned to just one object, in order to count the objects (e.g., Gelman, 1982 ). Halford et al. (2007) provide rough description of what complexity of concepts to expect for each age range, based on working-memory limits (see also Pascual-Leone & Johnson, 2011 ).

There also are individual differences within an age group in ability that affect how the materials are processed. For example, individuals lower in working memory may prefer to take in information using a verbatim, shallow, or surface processing strategy, rather than try to extract the gist (for one relevant investigation, albeit with mixed results, see Kyndt, Cascallar, & Dochy, 2012 ). The enjoyment of technological presentations may be greater in students with better abilities in the most relevant types of working memory (e.g., Garcia, Nussbaum, & Preiss, 2011 ). I would note that the educational enterprise requires that the teacher must decide whether it is best to allow the learner to use a favored strategy, which may be influenced by the student’s ability level, or whether it is possible in some cases to instill a more effective strategy even if it does not come naturally to the student.

Sweller and colleagues ( Sweller, 2011 ; Sweller, van Merrienboer, & Paas, 1998 ) have summarized a body of research literature and a theory about the role of cognitive load in learning and education. Their cognitive load theory is “a theory that emphasizes working memory constraints as determinants of instructional design effectiveness” ( Sweller et al., 1998 ). The theory distinguishes between an intrinsic cognitive load that comes from material to be learned and an extraneous cognitive load that should be kept small enough that the cognitive resources of the learner are not overly depleted by it. The theory is importantly placed in an evolutionary framework that I will not describe (though above I mentioned the theory’s incorporation of the distinction between biologically primary and secondary information). This theory has the advantage of being rather nuanced in that many ramifications of cognitive load are considered. With too high a cognitive load, one runs the risk of the student not being able to follow the presentation, whereas with too low a cognitive load, one runs the risk of insufficient engagement. In future, it might be possible to refine the predictions for classroom learning by combining cognitive load theory with theories of cognitive development, which make some specific predictions about how much capacity is present at a particular age in childhood (e.g., Halford et al., 2007 ). For further discussion of the theory as applied specifically to multimedia, see Schüler, Scheiter, and Genuchten (2011) . Issues arise as to how printed items are encoded (visually, verbally, or both) and how much the combination of verbal and visual codes in multimedia should be expected to tax a common, central cognitive resource and therefore interfere with one another, even when they are intended to be synergic. Both in cognitive psychology and in education, these are key issues currently under ongoing investigation.

An advantage of multimedia and computerized instruction is the possibility of adjusting the instruction to the student’s level. This might be done partly on the basis of success; if the student succeeds, the materials can be made more challenging whereas, if the student fails, the materials can be made easier. One potential pitfall to watch for is that, while some students will want to press slightly beyond their zone of comfort and will learn well, others will want an easy time, and may choose to learn less than they would be capable of learning. One way to cope with these issues is through computerized instruction, but with a heavy dose of personal monitoring and adjustment to make sure that the task is sufficiently motivating for every student.

One factor that makes it difficult to teach to the students effectively is that the working memory demands of language production do not always match the demands of the recipients’ language comprehension. Consequently, when one is speaking or writing for didactic purposes, one must be careful to consider not only one’s own working memory needs, but also those of the listener or reader. There are several obstacles in this regard. Slevc (2011) showed that speakers tend to blurt out what is most readily available in working memory. He used situations that were to be described verbally by the participant, e.g., A pirate gave a book to the monk . If one piece of information had already been presented, it was more likely to be described first. For example, if the monk had been presented already but not the book, the participant was more likely to phrase the description differently, as A pirate gave the monk a book . This assignment of priority to given information is generally appropriate, given that the speaker and listener (or writer and reader) share the same given information. In this case, though, Slevc shows that the tendency to describe given information first was diminished when the speaking participant was under a working memory load. In a didactic situation such as giving a lecture, it thus seems plausible that the memory load inherent in the situation (remembering and planning what one wants to say in the coming segments of a lecture) may cause the lecturer sometimes to use awkward grammatical structure. Moreover, as mentioned above, learning to speak or write well requires that one bear in mind possible difference between what one knows as the speaker (or writer) and what the listener (or reader) knows at key moments. For example, if one says, “Marconi was the inventor of the modern radio,” then, by the time the full topic of the sentence is known, the name is most likely no longer in the listener’s or student’s working memory. If, however, one says, “The modern radio was invented by a man named Marconi,” the context is set up first, making it easier to retain the name. Bearing in mind what the listener or reader knows and does not yet know is likely to be important both for educators in their own speaking and writing, and also in order to teach students how to speak and write effectively.

Working Memory Training

A much more controversial approach is to use training regimens to improve working memory, thereby improving performance on the educational learning tasks that require working memory (e.g., Klingberg, 2010 ). It is controversial partly because many people have spent a great deal of money purchasing such training programs before the scientific community has reached an agreement about the efficacy of such programs.

Doing working memory training studies is not easy. One needs a control group that is just as motivated by the task as the training group but without the working memory training aspect. The training task must be adaptive (with rewards for performance that continues to improve with training) and a non-adaptive control group does not adequately control arousal and motivation. Some task that is adaptive but involves long-term learning instead of working memory training may be adequate. Several large-scale reviews and studies have suggested that working memory training sometimes improves performance on the working memory task that is trained, but does not generalize to reasoning tasks that must rely on working memory (in adults, Redick et al., 2013 , and Shipstead, Redick, & Engle, 2012 ; in children, Melby-Lervåg & Hulme, 2013 ). In somewhat of a contrast, other reviews suggest that the training of executive functions (inhibiting irrelevant information, updating working memory, controlling attention, etc.) does extend at least to tasks that use similar processes ( Diamond & Lee, 2011 ) and some basically concur also for working memory ( Chein & Morrison, 2010 ). So there is an ongoing controversy, even among those who have written meta-analyses and reviews of research.

One might ask how it is possible to improve working memory without having the effect of improving performance on other tasks that rely on working memory. This can happen because there are potentially two ways in which training can improve task performance. First, working memory training theoretically might increase the function of a basic process, much as a muscle can be strengthened through practice. (Or at least, individuals might learn that through diligent exertion of their attention and effort, they can do better.) That is presumably the route hoped for in training of working memory or executive function. Second, though, it is possible for working memory training to result in the discovery of a strategy for completing the task that is better than the strategy used initially. This can improve performance on the task being trained, but the experience and the strategy learned may well be irrelevant to performance on other educational tasks, even those that rely on working memory. This route might be expected if, as I suspect, participants typically look for a way to solve a problem that is not very attention-demanding, unless the payoff is high.

If there is successful working memory training, another issue is whether training is capable of producing super-normal performance or whether it is mostly capable of rectifying deficiencies. By way of analogy, consider physical exercise. If a person is already walking 6 miles a day, there might be little benefit to the heart of adding aerobic exercise. Similarly, if a person is already highly engaged in the environment and using attention control often and effectively during the day, there might be little benefit to the brain of adding working memory exercises. It remains quite conceivable, though, that such exercises are beneficial to certain individuals who are under-utilizing working memory. Nevertheless, as Diamond and Lee (2011) points out, there might be social or emotional reasons why this is the case and such factors would need to be addressed along with, or in some cases instead of, working memory training per se.

Working Memory and the Ultimate Goals of Education

What is the difference between learning and education? This is a question that has long been asked (for a history of the early period of educational psychology in the United States, for example, see Hall, 2003 ). Do children learn better when they are fed the information intensively, or allowed to explore the material? Should all children be expected to learn the same material, or should children be separated into different tracks and taught the information that is thought to help them the most in their own most likely future walks of life?

A fundamental difference between learning and education, many would agree, is that education should facilitate the acquisition of skills that will promote continued learning after the student leaves school. Of course, after the student leaves school, a major difference is that there is no teacher to decide what is to be learned, or how. Therefore, what seems to be most important, many would agree, is critical thinking skills. There is some sentiment that these skills can be trained (although for an opposing view see Tricot & Sweller, in press ). For example, Halpern (1998 p. 449) suggests the following emphases for training critical thinking: “(a) a dispositional component to prepare learners for efforiful cognitive work, (b) instruction in the skills of critical thinking, (c) training in the structural aspects of problems and arguments to promote transcontextual transfer of critical-thinking skills, and (d) a metacognitive component that includes checking for accuracy and monitoring progress toward the goal.” Although I could find few well-controlled, peer-reviewed studies supporting the notion that it is possible to train critical thinking skills, optimistic evidence is beginning to roll in. For example, Shim and Walczak (2012) found that professors asking challenging questions resulted in more improvement in both subjective and objective measures of critical thinking. The objective measure required that students clarify, analyze, evaluate, and extend arguments, and increased 0.55 standardized units for every 1-unit increase in the rating of challenging questions asked. The gain was much stronger in students with high pretest scores in critical thinking. Halpern et al. (2012) have designed a computerized module to train critical thinking skills and obtained very encouraging initial results, with well-controlled training experiments in progress according to the report.

One can then ask, to what extent is the training of these higher-level skills dependent on the student’s working memory ability? The association is likely to be substantial, given the high correlation between working memory and reasoning ability even among normal adults ( Kyllonen & Christal, 1990 ; Süβ, Oberauer, Wittmann, Wilhelm, & Schulze, 2002 ). There is the possibility that training working memory will in some way improve reasoning and vice versa, though most would agree at this point that the case has not yet been completely made (e.g., Jaeggi & Buschkuehl, 2013 ; Shipstead et al., 2012 ).

A current interest of mine is to understand how fallacies in reasoning might be related to fallacies in working memory performance. There appear to be some similarities between the two. One of the best-known reasoning fallacies is confirmation bias. In a key example ( Wason & Shapiro, 1971 ) participants are given a set of cards laid on the table, each having a letter on one side and a number on the other, and are asked which cards must be turned over to assess a rule (e.g., If a card has a vowel on one side, it has an even number on the other side ). Participants get that they must turn over the cards that can either confirm or disconfirm the rule (in the example, the cards showing vowels). They often fail to realize that they must also turn over cards that can only disconfirm the rule. In the example, one must turn over cards with odd numbers because the rule is disconfirmed if any of those cards have a vowel on the other side. In contrast, cards that can only confirm the rule are irrelevant. (One should not turn over cards with even numbers because the rule is technically not disconfirmed no matter whether there is a consonant or vowel on the other side.)

Chen and Cowan (in press) found performance on a working memory task that closely resembles confirmation bias. In one procedure, a spatial array of letters was presented on each trial, followed by a set of all of the letters at the bottom of the screen and a single location marked; the task was to select the correct letter for the marked location. In another procedure, the spatial array of letters was followed by a single letter from the array at the bottom of the screen and all of the locations marked; the task was to select the correct location for the presented letter. When working memory does not happen to contain the probed item, these procedures allow the use of disconfirming information. In the first task, for example, a participant might reason as follows: The letters were K, R, Q, and L. I know the locations of only R and L and neither of them match the probed location. Therefore, I know that the answer must be K or Q and I will guess randomly between them . That would be comparable to using disconfirming evidence. The pattern of data, however, did not appear to indicate that kind of process. Instead, participants answered correctly if they knew the probed item and otherwise guessed randomly among all of the other choices, without using the process of elimination. A mathematical model that assumed the latter process showed near-perfect convergence in capacity between the procedures described above and the usual change-detection procedure. If we instead assumed a mathematical model of performance in which disconfirming evidence was used through the process of elimination, there was no such convergence between the procedures.

So in reasoning and in working memory, processing tends to be inefficient, and it remains to be seen whether it can be meaningfully improved in terms of eliminating confirmation bias. Perhaps people with insufficient working memory or intelligence will always be stuck in such inefficient reasoning and there is nothing we can do. Arguing against that pessimistic view, however, is the recent finding ( Stanovich, West, & Toplak, 2013 ) that the tendency to evaluate evidence more favorably when it agrees with one’s own view occurs across the board and is not correlated with intelligence, and presumably therefore not correlated with working memory either. One might be able to train individuals to make the best use of the working memory they have without worrying about increasing the basic capacity of working memory, either by training critical thinking skills (Halpern, 1989) or by instilling expertise ( Eriksson et al., 2004 ).

Working memory is the retention of a small amount of information in a readily accessible form, which facilitates planning, comprehension, reasoning, and problem-solving. When we talk of working memory, we often include not only the memory itself, but also the executive control skills that are used to manage information in working memory and the cognitive processing of information. Theoretically, there is still uncertainty about the basic limitations on working memory: are they limitations on concurrent holding capacity, mnemonic processing speed, duration of retention of information before it decays, or just the same sorts of interference properties that apply to long-term memory? While these basic issues are debated and empirical investigations continue, there is much greater agreement about what results are obtained in particular test circumstances; the results of working memory studies seem rather replicable, but small differences in method produce large differences in results, so that one cannot assume that a particular working memory finding is highly generalizable.

For learning and education, it is important to take into account the basic principles of cognitive development and cognitive psychology, adjusting the materials to the working memory capabilities of the learner. We are not yet at a point at which every task can be analyzed in advance in order to predict which tasks are doable with a particular working memory capability. It is possible, though, to monitor performance and keep in mind that failure could be due to working memory limitations, adjusting the presentation accordingly. Keeping in mind the limitations of working memory of listeners and readers could easily help to improve one’s lecturing and writing styles. I hope that awareness of working memory leads to a world in which we are all more tolerant of one another’s inability to understand perfectly, are more humble and less arrogant, and are better able to communicate, educate one another, and reach common ground.

Acknowledgment

This work was completed with support from NIH grant R01-HD21338.

• Andrews G, Halford GS. A cognitive complexity metric applied to cognitive development. Cognitive Psychology. 2002; 45 :153–219. [ PubMed ] [ Google Scholar ]
• Atkinson RC, Shiffrin RM. Human memory: A proposed system and its control processes. In: Spence KW, Spence JT, editors. The psychology of learning and motivation: Advances in research and theory. Vol. 2. New York: Academic Press; 1968. pp. 89–195. [ Google Scholar ]
• Baddeley AD. Working memory. Oxford: Clarendon Press; 1986. Oxford Psychology Series #11. [ Google Scholar ]
• Baddeley A. The episodic buffer: a new component of working memory? Trends in Cognitive Sciences. 2000; 4 :417–423. [ PubMed ] [ Google Scholar ]
• Baddeley AD, Gathercole SE, Papagno C. The phonological loop as a language learning device. Psychological Review. 1998; 105 :158–173. [ PubMed ] [ Google Scholar ]
• Barrouillet P, Gavens N, Vergauwe E, Gaillard V, Camos V. Working memory span development: A time-based resource-sharing model account. Developmental Psychology. 2009; 45 :477–490. [ PubMed ] [ Google Scholar ]
• Baddeley AD, Hitch G. In: Working memory. Bower GH, editor. Vol. 8. New York: Academic Press; 1974. pp. 47–89. The psychology of learning and motivation. [ Google Scholar ]
• Baddeley A, Papagno C, Vallar G. When long-term learning depends on short-term storage. Journal of Memory and Language. 1988; 27 :586–595. [ Google Scholar ]
• Barrouillet P, Portrat S, Camos V. On the law relating processing to storage in working memory. Psychological Review. 2011; 118 :175–192. [ PubMed ] [ Google Scholar ]
• Bauer PJ, Fivush R, editors. Handbook on the development of children’s memory. Wiley-Blackwell; in press. [ Google Scholar ]
• Bolton TL. The growth of memory in school children. American Journal of Psychology. 1892; 4 :362–380. [ Google Scholar ]
• Borst G, Niven E, Logie RH. Visual mental image generation does not overlap with visual short-term memory: A dual-task interference study. Memory & cognition. 2012; 40 :360–372. [ PubMed ] [ Google Scholar ]
• Broadbent DE. Perception and communication. New York: Pergamon Press; 1958. [ Google Scholar ]
• Brown GDA, Neath I, Chater N. A temporal ratio model of memory. Psychological Review. 2007; 114 :539–576. [ PubMed ] [ Google Scholar ]
• Burgess N, Hitch GJ. Memory for serial order: A network model of the phonological loop and its timing. Psychological Review. 1999; 106 :551–581. [ Google Scholar ]
• Camos V, Barrouillet P. Developmental change in working memory strategies: From passive maintenance to active refreshing. Developmental Psychology. 2011; 47 :898–904. [ PubMed ] [ Google Scholar ]
• Camos V, Mora G, Oberauer K. Adaptive choice between articulatory rehearsal and attentional refreshing in verbal working memory. Memory & Cognition. 2011; 39 :231–244. [ PubMed ] [ Google Scholar ]
• Case R, Kurland DM, Goldberg J. Operational efficiency and the growth of short-term memory span. Journal of Experimental Child Psychology. 1982; 33 :386–404. [ Google Scholar ]
• Chein J, Morrison A. Expanding the mind's workspace: training and transfer effects with a complex working memory span task. Psychonomic Bulletin & Review. 2010; 17 :193–199. [ PubMed ] [ Google Scholar ]
• Chen Z, Cowan N. Core verbal working memory capacity: The limit in words retained without covert articulation. Quarterly Journal of Experimental Psychology. 2009; 62 :1420–1429. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Chen Z, Cowan N. Working memory inefficiency: Minimal information is utilized in visual recognition tasks. Journal of Experimental Psychology: Learning, Memory, & Cognition. 2013; 39 :1449–1462. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Chi MTH. In: Knowledge structures and memory development. Siegler R, editor. Hillsdale, NJ: Erlbaum; 1978. Children(s thinking: What develops? [ Google Scholar ]
• Clark EV, Garnica OK. Is he coming or going? On the acquisition of deictic verbs. Journal of Verbal Learning and Verbal Behavior. 1974; 13 :559–572. [ Google Scholar ]
• Cocchini G, Logie RH, Della Sala S, MacPherson SE, Baddeley AD. Concurrent performance of two memory tasks: Evidence for domain-specific working memory systems. Memory & Cognition. 2002; 30 :1086–1095. [ PubMed ] [ Google Scholar ]
• Conrad R. Acoustic confusion in immediate memory. British Journal of Psychology. 1964; 55 :75–84. [ PubMed ] [ Google Scholar ]
• Courage ML, Cowan N, editors. The development of memory in infancy and childhood. Hove, U.K: Psychology Press; 2009. [ Google Scholar ]
• Cowan N. Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human information processing system. Psychological Bulletin. 1988; 104 :163–191. [ PubMed ] [ Google Scholar ]
• Cowan N. Verbal memory span and the timing of spoken recall. Journal of Memory and Language. 1992; 31 :668–684. [ Google Scholar ]
• Cowan N. Attention and memory: An integrated framework. New York: Oxford University Press; 1995. Oxford Psychology Series, No. 26. [ Google Scholar ]
• Cowan N. An embedded-processes model of working memory. In: Miyake A, Shah P, editors. Models of Working Memory: Mechanisms of active maintenance and executive control. Cambridge, U.K: Cambridge University Press; 1999. pp. 62–101. [ Google Scholar ]
• Cowan N. The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences. 2001; 24 :87–185. [ PubMed ] [ Google Scholar ]
• Cowan N. Working memory capacity. Hove, East Sussex, UK: Psychology Press; 2005. [ Google Scholar ]
• Cowan N. The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science. 2010; 19 :51–57. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N. The focus of attention as observed in visual working memory tasks: Making sense of competing claims. Neuropsychologia. 2011; 49 :1401–1406. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, AuBuchon AM, Gilchrist AL, Ricker TJ, Saults JS. Age differences in visual working memory capacity: Not based on encoding limitations. Developmental Science. 2011; 14 :1066–1074. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Donnell K, Saults JS. A list-length constraint on incidental item-to-item associations. Psychonomic Bulletin & Review. in press [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Li D, Moffitt A, Becker TM, Martin EA, Saults JS, Christ SE. A neural region of abstract working memory. Journal of Cognitive Neuroscience. 2011; 23 :2852–2863. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Morey CC. How can dual-task working memory retention limits be investigated? Psychological Science. 2007; 18 :686–688. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Morey CC, AuBuchon AM, Zwilling CE, Gilchrist AL. Seven-year-olds allocate attention like adults unless working memory is overloaded. Developmental Science. 2010; 13 :120–133. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Nugent LD, Elliott EM, Ponomarev I, Saults JS. The role of attention in the development of short-term memory: Age differences in the verbal span of apprehension. Child Development. 1999; 70 :1082–1097. [ PubMed ] [ Google Scholar ]
• Cowan N, Rouder JN, Blume CL, Saults JS. Models of verbal working memory capacity: What does it take to make them work? Psychological Review. 2012; 119 :480–499. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cowan N, Saults JS, Winterowd C, Sherk M. Enhancement of 4-year-old children's memory span for phonologically similar and dissimilar word lists. Journal of Experimental Child Psychology. 1991; 51 :30–52. [ PubMed ] [ Google Scholar ]
• Craik FIM, Watkins MJ. The role of rehearsal in short-term memory. Journal of Verbal Learning and Verbal Behavior. 1973; 12 :599–607. [ Google Scholar ]
• Daneman M, Carpenter PA. Individual differences in working memory and reading. Journal of Verbal Learning & Verbal Behavior. 1980; 19 :450–466. [ Google Scholar ]
• Diamond A, Lee K. Interventions shown to aid executive function development in children 4 to 12 years old. Science. 2011; 333 :959–964. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Ebbinghaus H. In: Memory: A contribution to experimental psychology. Ruger HA, Bussenius CE, translators. New York: Teachers College, Columbia University; 1885/1913. (Originally in German, Ueber das gedächtnis: Untersuchen zur experimentellen psychologie ) [ Google Scholar ]
• Engle RW, Tuholski SW, Laughlin JE, Conway ARA. Working memory, short term memory, and general fluid intelligence: A latent variable approach. Journal of Experimental Psychology: General. 1999; 128 :309–331. [ PubMed ] [ Google Scholar ]
• Ericsson KA, Chase WG, Faloon S. Acquisition of a memory skill. Science. 1980; 208 :1181–1182. [ PubMed ] [ Google Scholar ]
• Ericsson KA, Delaney PF, Weaver G, Mahadevan R. Uncovering the structure of a memorist’s superior basic-memory capacity. Cognitive Psychology. 2004; 49 :191–237. [ PubMed ] [ Google Scholar ]
• Ericsson KA, Kintsch W. Long-term working-memory. Psychological Review. 1995; 102 :211–245. [ PubMed ] [ Google Scholar ]
• Farrell S, Lewandowsky S. An endogenous distributed model of ordering in serial recall. Psychonomic Bulletin & Review. 2002; 9 :59–79. [ PubMed ] [ Google Scholar ]
• Flavell JH, Beach DH, Chinsky JM. Spontaneous verbal rehearsal in a memory task as a function of age. Child Development. 1966; 37 :283–299. [ PubMed ] [ Google Scholar ]
• Fougnie D, Marois R. What limits working memory capacity? Evidence for modality-specific sources to the simultaneous storage of visual and auditory arrays. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2011; 37 :1329–1341. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Garcia L, Nussbaum M, Preiss DD. Is the use of information and communication technology related to performance in working memory tasks? Evidence from seventh-grade students. Computers & Education. 2011; 57 :2068–2076. [ Google Scholar ]
• Garrity LI. An electromyographical study of subvocal speech and recall in preschool children. Developmental Psychology. 1975; 11 :274–281. [ Google Scholar ]
• Gathercole SE, Baddeley AD. Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study. Journal of Memory and Language. 1989; 28 :200–213. [ Google Scholar ]
• Gathercole SE, Baddeley AD. Phonological memory deficits in language disordered children: Is there a causal connection? Journal of Memory and Language. 1990; 29 :336–360. [ Google Scholar ]
• Gathercole SE, Lamont E, Alloway TP. Working memory in the classroom. In: Pickering SJ, editor. Working memory and education. San Diego: Academic Press; 2006. pp. 219–240. [ Google Scholar ]
• Gathercole SE, Pickering SJ, Ambridge B, Wearing H. The structure of working memory from 4 to 15 years of age. Developmental Psychology. 2004; 40 :177–190. [ PubMed ] [ Google Scholar ]
• Geary DC. An evolutionarily informed education science. Educational Psychologist. 2008; 43 :179–195. [ Google Scholar ]
• Gelman R. Accessing one-to-one correspondence: still another paper about conservation. British Journal of Psychology. 1982; 73 :209–220. [ Google Scholar ]
• Gilchrist AL, Cowan N, Naveh-Benjamin M. Investigating the childhood development of working memory using sentences: New evidence for the growth of chunk capacity. Journal of Experimental Child Psychology. 2009; 104 :252–265. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Halford GS, Cowan N, Andrews G. Separating cognitive capacity from knowledge: A new hypothesis. Trends in Cognitive Sciences. 2007; 11 :236–242. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Halford GS, Wilson WH, Phillips S. Processing capacity defined by relational complexity: Implications for comparative, developmental, and cognitive psychology. Behavioral and Brain Sciences. 1998; 21 :803–865. [ PubMed ] [ Google Scholar ]
• Hall VC. Educational psychology from 1890 to 1920. In: Zimmerman BJ, Schunk DH, editors. Educational psychology: A century of contributions. Mahwah, NJ: Erlbaum; 2003. [ Google Scholar ]
• Halpern DF. Teaching critical thinking for transfer across domains: Dispositions, skills, structure training, and metacogitive monitoring. American Psychologist. 1998; 53 :449–455. [ PubMed ] [ Google Scholar ]
• Halpern DF, Millis K, Graesser AC, Butler H, Forsyth C, Cai Z. Operation ARA: A computerized learning game that teaches critical thinking and scientific reasoning. Thinking Skills And Creativity. 2012; 7 :93–100. [ Google Scholar ]
• Hebb DO. Organization of behavior. New York: Wiley; 1949. [ Google Scholar ]
• Henry LA. The effects of word length and phonemic similarity in young children's short-term memory. Quarterly Journal of Experimental Psychology. 1991; 43A :35–52. [ Google Scholar ]
• Henson RNA. Short-term memory for serial order: The start-end model. Cognitive Psychology. 1998; 36 :73–137. [ PubMed ] [ Google Scholar ]
• Hulme C, Tordoff V. Working memory development: The effects of speech rate, word length, and acoustic similarity on serial recall. Journal of Experimental Child Psychology. 1989; 47 :72–87. [ Google Scholar ]
• Jaeggi SM, Buschkuehl M. Training working memory. In: Alloway TP, Alloway RG, editors. Working memory: The connected intelligence. NY: Psychology Press; 2013. [ Google Scholar ]
• James W. The principles of psychology. NY: Henry Holt; 1890. [ Google Scholar ]
• Jarrold C, Citroën R. Reevaluating key evidence for the development of rehearsal: Phonological similarity effects in children are subject to proportional scaling artifacts. Developmental Psychology. 2013; 49 :837–847. [ PubMed ] [ Google Scholar ]
• Jevons WS. The power of numerical discrimination. Nature. 1871; 3 :281–282. [ Google Scholar ]
• Jolicoeur P, Dell'Acqua R. The demonstration of short-term consolidation. Cognitive Psychology. 1998; 36 :138–202. [ PubMed ] [ Google Scholar ]
• Kail R. The development of memory in children. 3rd edition. NY: W.H.Freeman; 1990. [ Google Scholar ]
• Kail R, Salthouse TA. Processing speed as a mental capacity. Acta Psychologica. 1994; 86 :199–255. [ PubMed ] [ Google Scholar ]
• Kane MJ, Brown LH, McVay JC, Silvia PJ, Myin-Germeys I, Kwapil TR. For whom the mind wanders, and when: An experience-sampling study of working memory and executive control in daily life. Psychological Science. 2007; 18 :614–621. [ PubMed ] [ Google Scholar ]
• Kane MJ, Engle RW. Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General. 2003; 132 :47–70. [ PubMed ] [ Google Scholar ]
• Kane MJ, Hambrick DZ, Tuholski SW, Wilhelm O, Payne TW, Engle RW. The generality of workingmemory capacity: A latentvariable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General. 2004; 133 :189–217. [ PubMed ] [ Google Scholar ]
• Klingberg T. Training and plasticity of working memory. Trends in Cognitive Sciences. 2010; 14 :317–324. [ PubMed ] [ Google Scholar ]
• Kyllonen PC, Christal RE. Reasoning ability is (little more than) working-memory capacity?! Intelligence. 1990; 14 :389–433. [ Google Scholar ]
• Kyndt E, Cascallar E, Dochy F. Individual Differences in Working Memory Capacity and Attention, and Their Relationship with Students' Approaches to Learning. Higher Education: The International Journal Of Higher Education And Educational Planning. 2012; 64 :285–297. [ Google Scholar ]
• Lewandowsky S, Duncan M, Brown GDA. Time does not cause forgetting in short-term serial recall. Psychonomic Bulletin & Review. 2004; 11 :771–790. [ PubMed ] [ Google Scholar ]
• Locke J. An essay concerning human understanding. London: Thomas Bassett; 1690. [ Google Scholar ]
• Logie RH. The seven ages of working memory. In: Richardson JTE, Engle RW, Hasher L, Logie RH, Stoltzfus ER, Zacks RT, editors. Working memory and human cognition. New York: Oxford University Press; 1996. pp. 31–65. [ Google Scholar ]
• Logie RH, Van Der Meulen M. Fragmenting and integrating visuospatial working memory. In: Brockmole JR, editor. Representing the visual world in memory. Hove, U.K: Psychology Press; 2009. pp. 1–32. [ Google Scholar ]
• Luck SJ, Vogel EK. The capacity of visual working memory for features and conjunctions. Nature. 1997; 390 :279–281. [ PubMed ] [ Google Scholar ]
• Mandler G, Shebo BJ. Subitizing: An analysis of its component processes Journal of Experimental Psychology. General. 1982; 111 :1–22. [ PubMed ] [ Google Scholar ]
• Melby-Lervåg M, Hulme C. Is working memory training effective? A meta-analytic review. Developmental Psychology. 2013; 49 :270–291. [ PubMed ] [ Google Scholar ]
• Miller GA. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review. 1956; 63 :81–97. [ PubMed ] [ Google Scholar ]
• Miller GA, Galanter E, Pribram KH. Plans and the structure of behavior. New York: Holt, Rinehart and Winston, Inc; 1960. [ Google Scholar ]
• Morey CC, Bieler M. Visual short-term memory always requires attention. Psychonomic Bulletin & Review. 2013; 20 :163–170. [ PubMed ] [ Google Scholar ]
• Murdock BB. A distributed memory model for serial-order information. Psychological Review. 1983; 90 :316–338. [ Google Scholar ]
• Murdock BB, Walker KD. Modality effects in free recall. Journal of Verbal Leaning and Verbal Behavior. 1969; 8 :665–676. [ Google Scholar ]
• Nelson KJ. Variations in children's concepts by age and category. Child Development. 1974; 45 :577–584. [ PubMed ] [ Google Scholar ]
• Norman DA. Toward a theory of memory and attention. Psychological Review. 1968; 75 (6):522–536. [ Google Scholar ]
• Oberauer K, Lewandowsky S. Forgetting in immediate serial recall: decay, temporal distinctiveness, or interference? Psychological Review. 2008; 115 :544–576. [ PubMed ] [ Google Scholar ]
• Oberauer K, Lewandowsky S. Modeling working memory: A computational implementation of the time-based resource-sharing theory. Psychonomic Bulletin & Review. 2011; 18 :10–45. [ PubMed ] [ Google Scholar ]
• Oberauer K, Lewandowsky S, Farrell S, Jarrold C, Greaves M. Modeling working memory: An interference model of complex span. Psychonomic Bulletin & Review. 2012; 19 :779–819. [ PubMed ] [ Google Scholar ]
• Ornstein PA, Naus MJ. Rehearsal processes in children's memory. In: Ornstein PA, editor. Memory development in children. Hillsdale, NJ: Erlbaum; 1978. pp. 69–99. [ Google Scholar ]
• Paas F, Sweller J. An evolutionary upgrade of cognitive load theory: Using the human motor system and collaboration to support the learning of complex cognitive tasks. Educational Psychology Review. 2012; 24 :27–45. [ Google Scholar ]
• Pascual-Leone J, Johnson J. A developmental theory of mental attention: Its applications to measurement and task analysis. In: Barrouillet P, Gaillard V, editors. Cognitive development and working Memory: From neoPiagetian to cognitive approaches. Hove, UK: Psychology Press; 2011. pp. 13–46. [ Google Scholar ]
• Pascual-Leone J, Smith J. The encoding and decoding of symbols by children: A new experimental paradigm and a neo-Piagetian model. Journal of Experimental Child Psychology. 1969; 8 :328–355. [ Google Scholar ]
• Penney CG. Modality effects and the structure of short-term verbal memory. Memory & Cognition. 1989; 17 :398–422. [ PubMed ] [ Google Scholar ]
• Pickering SJ. Working memory and education. San Diego: Academic Press; 2006. [ Google Scholar ]
• Redick TS, Shipstead Z, Harrison TL, Hicks KL, Fried DE, Hambrick DZ, Kane MJ, Engle RW. No evidence of intelligence improvement after working memory training: A randomized, placebo-controlled study. Journal of Experimental Psychology: General. 2013; 142 :359–379. [ PubMed ] [ Google Scholar ]
• Ricker TJ, Cowan N. Loss of visual working memory within seconds: The combined use of refreshable and non-refreshable features. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2010; 36 :1355–1368. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Ricker TJ, Cowan N. Differences between presentation methods in working memory procedures: A matter of working memory consolidation. Journal of Experimental Psychology: Learning, Memory, and Cognition. in press [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Rouder JN, Morey RD, Cowan N, Zwilling CE, Morey CC, Pratte MS. An assessment of fixed-capacity models of visual working memory. Proceedings of the National Academy of Sciences (PNAS) 2008; 105 :5975–5979. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Sabol TJ, Pianta RC. Patterns of school readiness forecast achievement and socioemotional development at the end of elementary school. Child Development. 2012; 83 :282–299. [ PubMed ] [ Google Scholar ]
• Saltz E, Soller E, Sigel IE. The development of natural language concepts. Child Development. 1972; 43 :1191–1202. 1972. [ Google Scholar ]
• Schüler A, Scheiter K, Genuchten E. The Role of Working Memory in Multimedia Instruction: Is Working Memory Working During Learning from Text and Pictures? Educational Psychology Review. 2011; 23 :389–411. [ Google Scholar ]
• Shim W, Walczak K. The Impact of Faculty Teaching Practices on the Development of Students' Critical Thinking Skills. International Journal Of Teaching And Learning In Higher Education. 2012; 24 :16–30. [ Google Scholar ]
• Shipstead Z, Redick TS, Engle RW. Is working memory training effective? Psychological Bulletin. 2012; 138 :628–654. [ PubMed ] [ Google Scholar ]
• Slevc L. Saying What's on Your Mind: Working Memory Effects on Sentence Production. Journal Of Experimental Psychology: Learning, Memory, And Cognition. 2011; 37 :1503–1514. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Sokolov EN. Perception and the conditioned reflex. NY: Pergamon Press; 1963. [ Google Scholar ]
• Stanovich KE, West RF, Toplak ME. Myside Bias, Rational Thinking, and Intelligence. Current Directions in Psychological Science. 2013; 22 :259–264. [ Google Scholar ]
• Süβ HM, Oberauer K, Wittmann WW, Wilhelm O, Schulze R. Working-memory capacity explains reasoning ability—and a little bit more. Intelligence. 2002; 30 :261–288. [ Google Scholar ]
• Sweller J. Cognitive load theory. Psychology of Learning and Motivation. 2011; 55 :37–76. [ Google Scholar ]
• Sweller J, van Merrienboer JJG, Paas FGWC. Cognitive architecture and instructional design. Educational Psychology Review. 1998; 10 :251–296. [ Google Scholar ]
• Tan L, Ward G. A recency-based account of the primacy effect in free recall. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2000; 26 :1589–1625. [ PubMed ] [ Google Scholar ]
• Todd JJ, Marois R. Capacity limit of visual short-term memory in human posterior parietal cortex. Nature. 2004; 428 :751–754. [ PubMed ] [ Google Scholar ]
• Treisman AM. Contextual cues in selective listening. Quarterly Journal of Experimental Psychology. 1960; 12 :242–248. [ Google Scholar ]
• Tricot A, Sweller J. Domain-specific knowledge and why teaching generic skills does not work. Educational Psychology Review. in press [ Google Scholar ]
• Van Der Meulen M, Logie RH, Sala SD. Selective interference with image retention and generation: Evidence for the workspace model. The Quarterly Journal of Experimental Psychology. 2009; 62 :1568–1580. [ PubMed ] [ Google Scholar ]
• Vergauwe E, Barrouillet P, Camos V. Do mental processes share a domain-general resource? Psychological Science. 2010; 21 :384–390. [ PubMed ] [ Google Scholar ]
• Vogel EK, McCollough AW, Machizawa MG. Neural measures reveal individual differences in controlling access to working memory. Nature. 2005; 438 :500–503. [ PubMed ] [ Google Scholar ]
• Wason PC, Shapiro D. Natural and contrived experience in a reasoning problem. Quarterly Journal of Experimental Psychology. 1971; 23 :63–71. [ Google Scholar ]
• Wilding J. Over the top: Are there exceptions to the basic capacity limit? Behavioral and Brain Sciences. 2001; 24 :152–153. [ Google Scholar ]
• Wilhelm O, Engle RW, editors. Handbook of understanding and measuring intelligence. London: Sage; 2005. [ Google Scholar ]
• Wundt W. In: Lectures on human and animal psychology. Creighton JE, Titchener EB, editors. Bristol, U.K: Thoemmes Press; 1894/1998. Translated from the second German. [ Google Scholar ]
• Xu Y, Chun MM. Dissociable neural mechanisms supporting visual short-term memory for objects. Nature. 2006; 440 :91–95. [ PubMed ] [ Google Scholar ]
• Zhang W, Luck SJ. Sudden death and gradual decay in visual working memory. Psychological Science. 2009; 20 :423–428. [ PMC free article ] [ PubMed ] [ Google Scholar ]

Suggestions or feedback?

MIT News | Massachusetts Institute of Technology

• Machine learning
• Sustainability
• Black holes
• Classes and programs

Departments

• Aeronautics and Astronautics
• Brain and Cognitive Sciences
• Architecture
• Political Science
• Mechanical Engineering

Centers, Labs, & Programs

• Abdul Latif Jameel Poverty Action Lab (J-PAL)
• Picower Institute for Learning and Memory
• Lincoln Laboratory
• School of Architecture + Planning
• School of Engineering
• School of Humanities, Arts, and Social Sciences
• Sloan School of Management
• School of Science
• MIT Schwarzman College of Computing

New study reveals how brain waves control working memory

Press contact :, media download.

*Terms of Use:

Images for download on the MIT News office website are made available to non-commercial entities, press and the general public under a Creative Commons Attribution Non-Commercial No Derivatives license . You may not alter the images provided, other than to crop them to size. A credit line must be used when reproducing images; if one is not provided below, credit the images to "MIT."

Previous image Next image

MIT neuroscientists have found evidence that the brain’s ability to control what it’s thinking about relies on low-frequency brain waves known as beta rhythms.

In a memory task requiring information to be held in working memory for short periods of time, the MIT team found that the brain uses beta waves to consciously switch between different pieces of information. The findings support the researchers’ hypothesis that beta rhythms act as a gate that determines when information held in working memory is either read out or cleared out so we can think about something else.  

“The beta rhythm acts like a brake, controlling when to express information held in working memory and allow it to influence behavior,” says Mikael Lundqvist, a postdoc at MIT’s Picower Institute for Learning and Memory and the lead author of the study.

Earl Miller, the Picower Professor of Neuroscience at the Picower Institute and in the Department of Brain and Cognitive Sciences, is the senior author of the study, which appears in the Jan. 26 issue of Nature Communications .

Working in rhythm

There are millions of neurons in the brain, and each neuron produces its own electrical signals. These combined signals generate oscillations known as brain waves, which vary in frequency. In a 2016 study , Miller and Lundqvist found that gamma rhythms are associated with encoding and retrieving sensory information.

They also found that when gamma rhythms went up, beta rhythms went down, and vice versa. Previous work in their lab had shown that beta rhythms are associated with “top-down” information such as what the current goal is, how to achieve it, and what the rules of the task are.

All of this evidence led them to theorize that beta rhythms act as a control mechanism that determines what pieces of information are allowed to be read out from working memory — the brain function that allows control over conscious thought, Miller says.

“Working memory is the sketchpad of consciousness, and it is under our control. We choose what to think about,” he says. “You choose when to clear out working memory and choose when to forget about things. You can hold things in mind and wait to make a decision until you have more information.”

To test this hypothesis, the researchers recorded brain activity from the prefrontal cortex, which is the seat of working memory, in animals trained to perform a working memory task. The animals first saw one pair of objects, for example, A followed by B. Then they were shown a different pair and had to determine if it matched the first pair. A followed by B would be a match, but not B followed by A, or A followed by C. After this entire sequence, the animals released a bar if they determined that the two sequences matched.

The researchers found that brain activity varied depending on whether the two pairs matched or not. As an animal anticipated the beginning of the second sequence, it held the memory of object A, represented by gamma waves. If the next object seen was indeed A, beta waves then went up, which the researchers believe clears object A from working memory. Gamma waves then went up again, but this time the brain switched to holding information about object B, as this was now the relevant information to determine if the sequence matched.

However, if the first object shown was not a match for A, beta waves went way up, completely clearing out working memory, because the animal already knew that the sequence as a whole could not be a match.

“The interplay between beta and gamma acts exactly as you would expect a volitional control mechanism to act,” Miller says. “Beta is acting like a signal that gates access to working memory. It clears out working memory, and can act as a switch from one thought or item to another.”

A new model

Previous models of working memory proposed that information is held in mind by steady neuronal firing. The new study, in combination with their earlier work, supports the researchers’ new hypothesis that working memory is supported by brief episodes of spiking, which are controlled by beta rhythms.

“When we hold things in working memory (i.e. hold something ‘in mind’), we have the feeling that they are stable, like a light bulb that we’ve turned on to represent some thought. For a long time, neuroscientists have thought that this must mean that the way the brain represents these thoughts is through constant activity. This study shows that this isn’t the case — rather, our memories are blinking in and out of existence. Furthermore, each time a memory blinks on, it is riding on top of a wave of activity in the brain,” says Tim Buschman, an assistant professor of psychology at Princeton University who was not involved in the study.

Two other recent papers from Miller’s lab offer additional evidence for beta as a cognitive control mechanism.

In a study that recently appeared in the journal Neuron , they found similar patterns of interaction between beta and gamma rhythms in a different task involving assigning patterns of dots into categories. In cases where two patterns were easy to distinguish, gamma rhythms, carrying visual information, predominated during the identification. If the distinction task was more difficult, beta rhythms, carrying information about past experience with the categories, predominated.

In a recent paper published in the Proceedings of the National Academy of Sciences , Miller’s lab found that beta waves are produced by deep layers of the prefrontal cortex, and gamma rhythms are produced by superficial layers, which process sensory information. They also found that the beta waves were controlling the interaction of the two types of rhythms.

“When you find that kind of anatomical segregation and it’s in the infrastructure where you expect it to be, that adds a lot of weight to our hypothesis,” Miller says.

The researchers are now studying whether these types of rhythms control other brain functions such as attention. They also hope to study whether the interaction of beta and gamma rhythms explains why it is so difficult to hold more than a few pieces of information in mind at once.

“Eventually we’d like to see how these rhythms explain the limited capacity of working memory, why we can only hold a few thoughts in mind simultaneously, and what happens when you exceed capacity,” Miller says. “You have to have a mechanism that compensates for the fact that you overload your working memory and make decisions on which things are more important than others.”

The research was funded by the National Institute of Mental Health, the Office of Naval Research, and the Picower JFDP Fellowship.

Share this news article on:

Related links.

• Earl Miller
• Picower Institute
• Department of Brain and Cognitive Sciences

Related Topics

• Brain and cognitive sciences
• Neuroscience

Related Articles

Study: Rhythmic interactions between cortical layers underlie working memory

Brain waves reflect different types of learning

A new glimpse into working memory

How brain waves guide memory formation

Previous item Next item

More MIT News

Study: Early dark energy could resolve cosmology’s two biggest puzzles

Read full story →

3 Questions: The past, present, and future of sustainability science

Startup’s displays engineer light to generate immersive experiences without the headsets

3 Questions: What does innovation look like in the field of substance use disorder?

Celebrating student entrepreneurship at delta v’s 2024 Demo Day

MIT welcomes nine MLK Scholars for 2024-25

• More news on MIT News homepage →

Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA, USA

• Map (opens in new window)
• Events (opens in new window)
• People (opens in new window)
• Careers (opens in new window)
• Accessibility
• Social Media Hub
• MIT on Facebook
• MIT on YouTube
• MIT on Instagram

REVIEW article

Working memory from the psychological and neurosciences perspectives: a review.

• 1 Department of Neurosciences, School of Medical Sciences, Universiti Sains Malaysia, Kubang Kerian, Malaysia
• 2 Center for Neuroscience Services and Research, Universiti Sains Malaysia, Kubang Kerian, Malaysia

Since the concept of working memory was introduced over 50 years ago, different schools of thought have offered different definitions for working memory based on the various cognitive domains that it encompasses. The general consensus regarding working memory supports the idea that working memory is extensively involved in goal-directed behaviors in which information must be retained and manipulated to ensure successful task execution. Before the emergence of other competing models, the concept of working memory was described by the multicomponent working memory model proposed by Baddeley and Hitch. In the present article, the authors provide an overview of several working memory-relevant studies in order to harmonize the findings of working memory from the neurosciences and psychological standpoints, especially after citing evidence from past studies of healthy, aging, diseased, and/or lesioned brains. In particular, the theoretical framework behind working memory, in which the related domains that are considered to play a part in different frameworks (such as memory’s capacity limit and temporary storage) are presented and discussed. From the neuroscience perspective, it has been established that working memory activates the fronto-parietal brain regions, including the prefrontal, cingulate, and parietal cortices. Recent studies have subsequently implicated the roles of subcortical regions (such as the midbrain and cerebellum) in working memory. Aging also appears to have modulatory effects on working memory; age interactions with emotion, caffeine and hormones appear to affect working memory performances at the neurobiological level. Moreover, working memory deficits are apparent in older individuals, who are susceptible to cognitive deterioration. Another younger population with working memory impairment consists of those with mental, developmental, and/or neurological disorders such as major depressive disorder and others. A less coherent and organized neural pattern has been consistently reported in these disadvantaged groups. Working memory of patients with traumatic brain injury was similarly affected and shown to have unusual neural activity (hyper- or hypoactivation) as a general observation. Decoding the underlying neural mechanisms of working memory helps support the current theoretical understandings concerning working memory, and at the same time provides insights into rehabilitation programs that target working memory impairments from neurophysiological or psychological aspects.

Introduction

Working memory has fascinated scholars since its inception in the 1960’s ( Baddeley, 2010 ; D’Esposito and Postle, 2015 ). Indeed, more than a century of scientific studies revolving around memory in the fields of psychology, biology, or neuroscience have not completely agreed upon a unified categorization of memory, especially in terms of its functions and mechanisms ( Cowan, 2005 , 2008 ; Baddeley, 2010 ). From the coining of the term “memory” in the 1880’s by Hermann Ebbinghaus, to the distinction made between primary and secondary memory by William James in 1890, and to the now widely accepted and used categorizations of memory that include: short-term, long-term, and working memories, studies that have tried to decode and understand this abstract concept called memory have been extensive ( Cowan, 2005 , 2008 ). Short and long-term memory suggest that the difference between the two lies in the period that the encoded information is retained. Other than that, long-term memory has been unanimously understood as a huge reserve of knowledge about past events, and its existence in a functioning human being is without dispute ( Cowan, 2008 ). Further categorizations of long-term memory include several categories: (1) episodic; (2) semantic; (3) Pavlovian; and (4) procedural memory ( Humphreys et al., 1989 ). For example, understanding and using language in reading and writing demonstrates long-term storage of semantics. Meanwhile, short-term memory was defined as temporarily accessible information that has a limited storage time ( Cowan, 2008 ). Holding a string of meaningless numbers in the mind for brief delays reflects this short-term component of memory. Thus, the concept of working memory that shares similarities with short-term memory but attempts to address the oversimplification of short-term memory by introducing the role of information manipulation has emerged ( Baddeley, 2012 ). This article seeks to present an up-to-date introductory overview of the realm of working memory by outlining several working memory studies from the psychological and neurosciences perspectives in an effort to refine and unite the scientific knowledge concerning working memory.

The Multicomponent Working Memory Model

When one describes working memory, the multicomponent working memory model is undeniably one of the most prominent working memory models that is widely cited in literatures ( Baars and Franklin, 2003 ; Cowan, 2005 ; Chein et al., 2011 ; Ashkenazi et al., 2013 ; D’Esposito and Postle, 2015 ; Kim et al., 2015 ). Baddeley and Hitch (1974) proposed a working memory model that revolutionized the rigid and dichotomous view of memory as being short or long-term, although the term “working memory” was first introduced by Miller et al. (1960) . The working memory model posited that as opposed to the simplistic functions of short-term memory in providing short-term storage of information, working memory is a multicomponent system that manipulates information storage for greater and more complex cognitive utility ( Baddeley and Hitch, 1974 ; Baddeley, 1996 , 2000b ). The three subcomponents involved are phonological loop (or the verbal working memory), visuospatial sketchpad (the visual-spatial working memory), and the central executive which involves the attentional control system ( Baddeley and Hitch, 1974 ; Baddeley, 2000b ). It was not until 2000 that another component termed “episodic buffer” was introduced into this working memory model ( Baddeley, 2000a ). Episodic buffer was regarded as a temporary storage system that modulates and integrates different sensory information ( Baddeley, 2000a ). In short, the central executive functions as the “control center” that oversees manipulation, recall, and processing of information (non-verbal or verbal) for meaningful functions such as decision-making, problem-solving or even manuscript writing. In Baddeley and Hitch (1974) ’s well-cited paper, information received during the engagement of working memory can also be transferred to long-term storage. Instead of seeing working memory as merely an extension and a useful version of short-term memory, it appears to be more closely related to activated long-term memory, as suggested by Cowan (2005 , 2008 ), who emphasized the role of attention in working memory; his conjectures were later supported by Baddeley (2010) . Following this, the current development of the multicomponent working memory model could be retrieved from Baddeley’s article titled “Working Memory” published in Current Biology , in Figure 2 ( Baddeley, 2010 ).

An Embedded-Processes Model of Working Memory

Notwithstanding the widespread use of the multicomponent working memory model, Cowan (1999 , 2005 ) proposed the embedded-processes model that highlights the roles of long-term memory and attention in facilitating working memory functioning. Arguing that the Baddeley and Hitch (1974) model simplified perceptual processing of information presentation to the working memory store without considering the focus of attention to the stimuli presented, Cowan (2005 , 2010 ) stressed the pivotal and central roles of working memory capacity for understanding the working memory concept. According to Cowan (2008) , working memory can be conceptualized as a short-term storage component with a capacity limit that is heavily dependent on attention and other central executive processes that make use of stored information or that interact with long-term memory. The relationships between short-term, long-term, and working memory could be presented in a hierarchical manner whereby in the domain of long-term memory, there exists an intermediate subset of activated long-term memory (also the short-term storage component) and working memory belongs to the subset of activated long-term memory that is being attended to ( Cowan, 1999 , 2008 ). An illustration of Cowan’s theoretical framework on working memory can be traced back to Figure 1 in his paper titled “What are the differences between long-term, short-term, and working memory?” published in Progress in Brain Research ( Cowan, 2008 ).

Alternative Models

Cowan’s theoretical framework toward working memory is consistent with Engle (2002) ’s view, in which it was posited that working memory capacity is comparable to directed or held attention information inhibition. Indeed, in their classic study on reading span and reading comprehension, Daneman and Carpenter (1980) demonstrated that working memory capacity, which was believed to be reflected by the reading span task, strongly correlated with various comprehension tests. Surely, recent and continual growth in the memory field has also demonstrated the development of other models such as the time-based resource-sharing model proposed by several researchers ( Barrouillet et al., 2004 , 2009 ; Barrouillet and Camos, 2007 ). This model similarly demonstrated that cognitive load and working memory capacity that were so often discussed by working memory researchers were mainly a product of attention that one receives to allocate to tasks at hand ( Barrouillet et al., 2004 , 2009 ; Barrouillet and Camos, 2007 ). In fact, the allocated cognitive resources for a task (such as provided attention) and the duration of such allocation dictated the likelihood of success in performing the tasks ( Barrouillet et al., 2004 , 2009 ; Barrouillet and Camos, 2007 ). This further highlighted the significance of working memory in comparison with short-term memory in that, although information retained during working memory is not as long-lasting as long-term memory, it is not the same and deviates from short-term memory for it involves higher-order processing and executive cognitive controls that are not observed in short-term memory. A more detailed presentation of other relevant working memory models that shared similar foundations with Cowan’s and emphasized the roles of long-term memory can be found in the review article by ( D’Esposito and Postle, 2015 ).

In addition, in order to understand and compare similarities and disparities in different proposed models, about 20 years ago, Miyake and Shah (1999) suggested theoretical questions to authors of different models in their book on working memory models. The answers to these questions and presentations of models by these authors gave rise to a comprehensive definition of working memory proposed by Miyake and Shah (1999 , p. 450), “working memory is those mechanisms or processes that are involved in the control, regulation, and active maintenance of task-relevant information in the service of complex cognition, including novel as well as familiar, skilled tasks. It consists of a set of processes and mechanisms and is not a fixed ‘place’ or ‘box’ in the cognitive architecture. It is not a completely unitary system in the sense that it involves multiple representational codes and/or different subsystems. Its capacity limits reflect multiple factors and may even be an emergent property of the multiple processes and mechanisms involved. Working memory is closely linked to LTM, and its contents consist primarily of currently activated LTM representations, but can also extend to LTM representations that are closely linked to activated retrieval cues and, hence, can be quickly activated.” That said, in spite of the variability and differences that have been observed following the rapid expansion of working memory understanding and its range of models since the inception of the multicomponent working memory model, it is worth highlighting that the roles of executive processes involved in working memory are indisputable, irrespective of whether different components exist. Such notion is well-supported as Miyake and Shah, at the time of documenting the volume back in the 1990’s, similarly noted that the mechanisms of executive control were being heavily investigated and emphasized ( Miyake and Shah, 1999 ). In particular, several domains of working memory such as the focus of attention ( Cowan, 1999 , 2008 ), inhibitory controls ( Engle and Kane, 2004 ), maintenance, manipulation, and updating of information ( Baddeley, 2000a , 2010 ), capacity limits ( Cowan, 2005 ), and episodic buffer ( Baddeley, 2000a ) were executive processes that relied on executive control efficacy (see also Miyake and Shah, 1999 ; Barrouillet et al., 2004 ; D’Esposito and Postle, 2015 ).

The Neuroscience Perspective

Following such cognitive conceptualization of working memory developed more than four decades ago, numerous studies have intended to tackle this fascinating working memory using various means such as decoding its existence at the neuronal level and/or proposing different theoretical models in terms of neuronal activity or brain activation patterns. Table 1 offers the summarized findings of these literatures. From the cognitive neuroscientific standpoint, for example, the verbal and visual-spatial working memories were examined separately, and the distinction between the two forms was documented through studies of patients with overt impairment in short-term storage for different verbal or visual tasks ( Baddeley, 2000b ). Based on these findings, associations or dissociations with the different systems of working memory (such as phonological loops and visuospatial sketchpad) were then made ( Baddeley, 2000b ). It has been established that verbal and acoustic information activates Broca’s and Wernicke’s areas while visuospatial information is represented in the right hemisphere ( Baddeley, 2000b ). Not surprisingly, many supporting research studies have pointed to the fronto-parietal network involving the dorsolateral prefrontal cortex (DLPFC), the anterior cingulate cortex (ACC), and the parietal cortex (PAR) as the working memory neural network ( Osaka et al., 2003 ; Owen et al., 2005 ; Chein et al., 2011 ; Kim et al., 2015 ). More precisely, the DLPFC has been largely implicated in tasks demanding executive control such as those requiring integration of information for decision-making ( Kim et al., 2015 ; Jimura et al., 2017 ), maintenance and manipulation/retrieval of stored information or relating to taxing loads (such as capacity limit) ( Osaka et al., 2003 ; Moore et al., 2013 ; Vartanian et al., 2013 ; Rodriguez Merzagora et al., 2014 ), and information updating ( Murty et al., 2011 ). Meanwhile, the ACC has been shown to act as an “attention controller” that evaluates the needs for adjustment and adaptation of received information based on task demands ( Osaka et al., 2003 ), and the PAR has been regarded as the “workspace” for sensory or perceptual processing ( Owen et al., 2005 ; Andersen and Cui, 2009 ). Figure 1 attempted to translate the theoretical formulation of the multicomponent working memory model ( Baddeley, 2010 ) to specific regions in the human brain. It is, however, to be acknowledged that the current neuroscientific understanding on working memory adopted that working memory, like other cognitive systems, involves the functional integration of the brain as a whole; and to clearly delineate its roles into multiple components with only a few regions serving as specific buffers was deemed impractical ( D’Esposito and Postle, 2015 ). Nonetheless, depicting the multicomponent working memory model in the brain offers a glimpse into the functional segregation of working memory.

TABLE 1. Working memory (WM) studies in the healthy brain.

FIGURE 1. A simplified depiction (adapted from the multicomponent working memory model by Baddeley, 2010 ) as implicated in the brain, in which the central executive assumes the role to exert control and oversee the manipulation of incoming information for intended execution. ACC, Anterior cingulate cortex.

Further investigation has recently revealed that other than the generally informed cortical structures involved in verbal working memory, basal ganglia, which lies in the subcortical layer, plays a role too ( Moore et al., 2013 ). Particularly, the caudate and thalamus were activated during task encoding, and the medial thalamus during the maintenance phase, while recorded activity in the fronto-parietal network, which includes the DLPFC and the parietal lobules, was observed only during retrieval ( Moore et al., 2013 ). These findings support the notion that the basal ganglia functions to enhance focusing on a target while at the same time suppressing irrelevant distractors during verbal working memory tasks, which is especially crucial at the encoding phase ( Moore et al., 2013 ). Besides, a study conducted on mice yielded a similar conclusion in which the mediodorsal thalamus aided the medial prefrontal cortex in the maintenance of working memory ( Bolkan et al., 2017 ). In another study by Murty et al. (2011) in which information updating, which is one of the important aspects of working memory, was investigated, the midbrain including the substantia nigra/ventral tegmental area and caudate was activated together with DLPFC and other parietal regions. Taken together, these studies indicated that brain activation of working memory are not only limited to the cortical layer ( Murty et al., 2011 ; Moore et al., 2013 ). In fact, studies on cerebellar lesions subsequently discovered that patients suffered from impairments in attention-related working memory or executive functions, suggesting that in spite of the motor functions widely attributed to the cerebellum, the cerebellum is also involved in higher-order cognitive functions including working memory ( Gottwald et al., 2004 ; Ziemus et al., 2007 ).

Shifting the attention to the neuronal network involved in working memory, effective connectivity analysis during engagement of a working memory task reinforced the idea that the DLPFC, PAR and ACC belong to the working memory circuitry, and bidirectional endogenous connections between all these regions were observed in which the left and right PAR were the modeled input regions ( Dima et al., 2014 ) (refer to Supplementary Figure 1 in Dima et al., 2014 ). Effective connectivity describes the attempt to model causal influence of neuronal connections in order to better understand the hidden neuronal states underlying detected neuronal responses ( Friston et al., 2013 ). Another similar study of working memory using an effective connectivity analysis that involved more brain regions, including the bilateral middle frontal gyrus (MFG), ACC, inferior frontal cortex (IFC), and posterior parietal cortex (PPC) established the modulatory effect of working memory load in this fronto-parietal network with memory delay as the driving input to the bilateral PPC ( Ma et al., 2012 ) (refer to Figure 1 in Ma et al., 2012 ).

Moving away from brain regions activated but toward the in-depth neurobiological side of working memory, it has long been understood that the limited capacity of working memory and its transient nature, which are considered two of the defining characteristics of working memory, indicate the role of persistent neuronal firing (see Review Article by D’Esposito and Postle, 2015 ; Zylberberg and Strowbridge, 2017 ; see also Silvanto, 2017 ), that is, continuous action potentials are generated in neurons along the neural network. However, this view was challenged when activity-silent synaptic mechanisms were found to also be involved ( Mongillo et al., 2008 ; Rose et al., 2016 ; see also Silvanto, 2017 ). Instead of holding relevant information through heightened and persistent neuronal firing, residual calcium at the presynaptic terminals was suggested to have mediated the working memory process ( Mongillo et al., 2008 ). This synaptic theory was further supported when TMS application produced a reactivation effect of past information that was not needed or attended at the conscious level, hence the TMS application facilitated working memory efficacy ( Rose et al., 2016 ). As it happens, this provided evidence from the neurobiological viewpoint to support Cowan’s theorized idea of “activated long-term memory” being a feature of working memory as non-cued past items in working memory that were assumed to be no longer accessible were actually stored in a latent state and could be brought back into consciousness. However, the researchers cautioned the use of the term “activated long-term memory” and opted for “prioritized long-term memory” because these unattended items maintained in working memory seemed to employ a different mechanism than items that were dropped from working memory ( Rose et al., 2016 ). Other than the synaptic theory, the spiking working memory model proposed by Fiebig and Lansner (2017) that borrowed the concept from fast Hebbian plasticity similarly disagreed with persistent neuronal activity and demonstrated that working memory processes were instead manifested in discrete oscillatory bursts.

Age and Working Memory

Nevertheless, having established a clear working memory circuitry in the brain, differences in brain activations, neural patterns or working memory performances are still apparent in different study groups, especially in those with diseased or aging brains. For a start, it is well understood that working memory declines with age ( Hedden and Gabrieli, 2004 ; Ziaei et al., 2017 ). Hence, older participants are expected to perform poorer on a working memory task when making comparison with relatively younger task takers. In fact, it was reported that decreases in cortical surface area in the frontal lobe of the right hemisphere was associated with poorer performers ( Nissim et al., 2017 ). In their study, healthy (those without mild cognitive impairments [MCI] or neurodegenerative diseases such as dementia or Alzheimer’s) elderly people with an average age of 70 took the n-back working memory task while magnetic resonance imaging (MRI) scans were obtained from them ( Nissim et al., 2017 ). The outcomes exhibited that a decrease in cortical surface areas in the superior frontal gyrus, pars opercularis of the inferior frontal gyrus, and medial orbital frontal gyrus that was lateralized to the right hemisphere, was significantly detected among low performers, implying an association between loss of brain structural integrity and working memory performance ( Nissim et al., 2017 ). There was no observed significant decline in cortical thickness of the studied brains, which is assumed to implicate neurodegenerative tissue loss ( Nissim et al., 2017 ).

Moreover, another extensive study that examined cognitive functions of participants across the lifespan using functional magnetic resonance imaging (fMRI) reported that the right lateralized fronto-parietal regions in addition to the ventromedial prefrontal cortex (VMPFC), posterior cingulate cortex, and left angular and middle frontal gyri (the default mode regions) in older adults showed reduced modulation of task difficulty, which was reflective of poorer task performance ( Rieck et al., 2017 ). In particular, older-age adults (55–69 years) exhibited diminished brain activations (positive modulation) as compared to middle-age adults (35–54 years) with increasing task difficulty, whereas lesser deactivation (negative modulation) was observed between the transition from younger adults (20–34 years) to middle-age adults ( Rieck et al., 2017 ). This provided insights on cognitive function differences during an individual’s lifespan at the neurobiological level, which hinted at the reduced ability or efficacy of the brain to modulate functional regions to increased difficulty as one grows old ( Rieck et al., 2017 ). As a matter of fact, such an opinion was in line with the Compensation-Related Utilization of Neural Circuits Hypothesis (CRUNCH) proposed by Reuter-Lorenz and Cappell (2008) . The CRUNCH likewise agreed upon reduced neural efficiency in older adults and contended that age-associated cognitive decline brought over-activation as a compensatory mechanism; yet, a shift would occur as task loads increase and under-activation would then be expected because older adults with relatively lesser cognitive resources would max out their ‘cognitive reserve’ sooner than younger adults ( Reuter-Lorenz and Park, 2010 ; Schneider-Garces et al., 2010 ).

In addition to those findings, emotional distractors presented during a working memory task were shown to alter or affect task performance in older adults ( Oren et al., 2017 ; Ziaei et al., 2017 ). Based on the study by Oren et al. (2017) who utilized the n-back task paired with emotional distractors with neutral or negative valence in the background, negative distractors with low load (such as 1-back) resulted in shorter response time (RT) in the older participants ( M age = 71.8), although their responses were not significantly more accurate when neutral distractors were shown. Also, lesser activations in the bilateral MFG, VMPFC, and left PAR were reported in the old-age group during negative low load condition. This finding subsequently demonstrated the results of emotional effects on working memory performance in older adults ( Oren et al., 2017 ). Further functional connectivity analyses revealed that the amygdala, the region well-known to be involved in emotional processing, was deactivated and displayed similar strength in functional connectivity regardless of emotional or load conditions in the old-age group ( Oren et al., 2017 ). This finding went in the opposite direction of that observed in the younger group in which the amygdala was strongly activated with less functional connections to the bilateral MFG and left PAR ( Oren et al., 2017 ). This might explain the shorter reported RT, which was an indication of improved working memory performance, during the emotional working memory task in the older adults as their amygdala activation was suppressed as compared to the younger adults ( Oren et al., 2017 ).

Interestingly, a contrasting neural connection outcome was reported in the study by Ziaei et al. (2017) in which differential functional networks relating to emotional working memory task were employed by the two studied groups: (1) younger ( M age = 22.6) and (2) older ( M age = 68.2) adults. In the study, emotional distractors with positive, neutral, and negative valence were presented during a visual working memory task and older adults were reported to adopt two distinct networks involving the VMPFC to encode and process positive and negative distractors while younger adults engaged only one neural pathway ( Ziaei et al., 2017 ). The role of amygdala engagement in processing only negative items in the younger adults, but both negative and positive distractors in the older adults, could be reflective of the older adults’ better ability at regulating negative emotions which might subsequently provide a better platform for monitoring working memory performance and efficacy as compared to their younger counterparts ( Ziaei et al., 2017 ). This study’s findings contradict those by Oren et al. (2017) in which the amygdala was found to play a bigger role in emotional working memory tasks among older participants as opposed to being suppressed as reported by Oren et al. (2017) . Nonetheless, after overlooking the underlying neural mechanism relating to emotional distractors, it was still agreed that effective emotional processing sustained working memory performance among older/elderly people ( Oren et al., 2017 ; Ziaei et al., 2017 ).

Aside from the interaction effect between emotion and aging on working memory, the impact of caffeine was also investigated among elders susceptible to age-related cognitive decline; and those reporting subtle cognitive deterioration 18-months after baseline measurement showed less marked effects of caffeine in the right hemisphere, unlike those with either intact cognitive ability or MCI ( Haller et al., 2017 ). It was concluded that while caffeine’s effects were more pronounced in MCI participants, elders in the early stages of cognitive decline displayed diminished sensitivity to caffeine after being tested with the n-back task during fMRI acquisition ( Haller et al., 2017 ). It is, however, to be noted that the working memory performance of those displaying minimal cognitive deterioration was maintained even though their brain imaging uncovered weaker brain activation in a more restricted area ( Haller et al., 2017 ). Of great interest, such results might present a useful brain-based marker that can be used to identify possible age-related cognitive decline.

Similar findings that demonstrated more pronounced effects of caffeine on elderly participants were reported in an older study, whereas older participants in the age range of 50–65 years old exhibited better working memory performance that offset the cognitive decline observed in those with no caffeine consumption, in addition to displaying shorter reaction times and better motor speeds than observed in those without caffeine ( Rees et al., 1999 ). Animal studies using mice showed replication of these results in mutated mice models of Alzheimer’s disease or older albino mice, both possibly due to the reported results of reduced amyloid production or brain-derived neurotrophic factor and tyrosine-kinase receptor. These mice performed significantly better after caffeine treatment in tasks that supposedly tapped into working memory or cognitive functions ( Arendash et al., 2006 ). Such direct effects of caffeine on working memory in relation to age was further supported by neuroimaging studies ( Haller et al., 2013 ; Klaassen et al., 2013 ). fMRI uncovered increased brain activation in regions or networks of working memory, including the fronto-parietal network or the prefrontal cortex in old-aged ( Haller et al., 2013 ) or middle-aged adults ( Klaassen et al., 2013 ), even though the behavioral measures of working memory did not differ. Taken together, these outcomes offered insight at the neurobiological level in which caffeine acts as a psychoactive agent that introduces changes and alters the aging brain’s biological environment that explicit behavioral testing might fail to capture due to performance maintenance ( Haller et al., 2013 , 2017 ; Klaassen et al., 2013 ).

With respect to physiological effects on cognitive functions (such as effects of caffeine on brain physiology), estradiol, the primary female sex hormone that regulates menstrual cycles, was found to also modulate working memory by engaging different brain activity patterns during different phases of the menstrual cycle ( Joseph et al., 2012 ). The late follicular (LF) phase of the menstrual cycle, characterized by high estradiol levels, was shown to recruit more of the right hemisphere that was associated with improved working memory performance than did the early follicular (EF) phase, which has lower estradiol levels although overall, the direct association between estradiol levels and working memory was inconclusive ( Joseph et al., 2012 ). The finding that estradiol levels modified brain recruitment patterns at the neurobiological level, which could indirectly affect working memory performance, presents implications that working memory impairment reported in post-menopausal women (older aged women) could indicate a link with estradiol loss ( Joseph et al., 2012 ). In 2000, post-menopausal women undergoing hormone replacement therapy, specifically estrogen, were found to have better working memory performance in comparison with women who took estrogen and progestin or women who did not receive the therapy ( Duff and Hampson, 2000 ). Yet, interestingly, a study by Janowsky et al. (2000) showed that testosterone supplementation counteracted age-related working memory decline in older males, but a similar effect was not detected in older females who were supplemented with estrogen. A relatively recent paper might have provided the explanation to such contradicting outcomes ( Schöning et al., 2007 ). As demonstrated in the study using fMRI, the nature of the task (such as verbal or visual-spatial) might have played a role as a higher level of testosterone (in males) correlated with activations of the left inferior parietal cortex, which was deemed a key region in spatial processing that subsequently brought on better performance in a mental-rotation task. In contrast, significant correlation between estradiol and other cortical activations in females in the midluteal phase, who had higher estradiol levels, did not result in better performance of the task compared to women in the EF phase or men ( Schöning et al., 2007 ). Nonetheless, it remains premature to conclude that age-related cognitive decline was a result of hormonal (estradiol or testosterone) fluctuations although hormones might have modulated the effect of aging on working memory.

Other than the presented interaction effects of age and emotions, caffeine, and hormones, other studies looked at working memory training in the older population in order to investigate working memory malleability in the aging brain. Findings of improved performance for the same working memory task after training were consistent across studies ( Dahlin et al., 2008 ; Borella et al., 2017 ; Guye and von Bastian, 2017 ; Heinzel et al., 2017 ). Such positive results demonstrated effective training gains regardless of age difference that could even be maintained until 18 months later ( Dahlin et al., 2008 ) even though the transfer effects of such training to other working memory tasks need to be further elucidated as strong evidence of transfer with medium to large effect size is lacking ( Dahlin et al., 2008 ; Guye and von Bastian, 2017 ; Heinzel et al., 2017 ; see also Karbach and Verhaeghen, 2014 ). The studies showcasing the effectiveness of working memory training presented a useful cognitive intervention that could partially stall or delay cognitive decline. Table 2 presents an overview of the age-related working memory studies.

TABLE 2. Working memory (WM) studies in relation to age.

The Diseased Brain and Working Memory

Age is not the only factor influencing working memory. In recent studies, working memory deficits in populations with mental or neurological disorders were also being investigated (see Table 3 ). Having identified that the working memory circuitry involves the fronto-parietal region, especially the prefrontal and parietal cortices, in a healthy functioning brain, targeting these areas in order to understand how working memory is affected in a diseased brain might provide an explanation for the underlying deficits observed at the behavioral level. For example, it was found that individuals with generalized or social anxiety disorder exhibited reduced DLPFC activation that translated to poorer n-back task performance in terms of accuracy and RT when compared with the controls ( Balderston et al., 2017 ). Also, VMPFC and ACC, representing the default mode network (DMN), were less inhibited in these individuals, indicating that cognitive resources might have been divided and resulted in working memory deficits due to the failure to disengage attention from persistent anxiety-related thoughts ( Balderston et al., 2017 ). Similar speculation can be made about individuals with schizophrenia. Observed working memory deficits might be traced back to impairments in the neural networks that govern attentional-control and information manipulation and maintenance ( Grot et al., 2017 ). The participants performed a working memory binding task, whereby they had to make sure that the word-ellipse pairs presented during the retrieval phase were identical to those in the encoding phase in terms of location and verbal information; results concluded that participants with schizophrenia had an overall poorer performance compared to healthy controls when they were asked to actively bind verbal and spatial information ( Grot et al., 2017 ). This was reflected in the diminished activation in the schizophrenia group’s ventrolateral prefrontal cortex and the PPC that were said to play a role in manipulation and reorganization of information during encoding and maintenance of information after encoding ( Grot et al., 2017 ).

TABLE 3. Working memory (WM) studies in the diseased brain.

In addition, patients with major depressive disorder (MDD) displayed weaker performance in the working memory updating domain in which information manipulation was needed when completing a visual working memory task ( Le et al., 2017 ). The working memory task employed in the study was a delayed recognition task that required participants to remember and recognize the faces or scenes as informed after stimuli presentation while undergoing fMRI scan ( Le et al., 2017 ). Subsequent functional connectivity analyses revealed that the fusiform face area (FFA), parahippocampal place area (PPA), and left MFG showed aberrant activity in the MDD group as compared to the control group ( Le et al., 2017 ). These brain regions are known to be the visual association area and the control center of working memory and have been implicated in visual working memory updating in healthy adults ( Le et al., 2017 ). Therefore, altered visual cortical functions and load-related activation in the prefrontal cortex in the MDD group implied that the cognitive control for visual information processing and updating might be impaired at the input or control level, which could have ultimately played a part in the depressive symptoms ( Le et al., 2017 ).

Similarly, during a verbal delayed match to sample task that asked participants to sub-articulatorly rehearse presented target letters for subsequent letter-matching, individuals with bipolar affective disorder displayed aberrant neural interactions between the right amygdala, which is part of the limbic system implicated in emotional processing as previously described, and ipsilateral cortical regions often concerned with verbal working memory, pointing out that the cortico-amygdalar connectivity was disrupted, which led to verbal working memory deficits ( Stegmayer et al., 2015 ). As an attempt to gather insights into previously reported hyperactivation in the amygdala in bipolar affective disorder during an articulatory working memory task, functional connectivity analyses revealed that negative functional interactions seen in healthy controls were not replicated in patients with bipolar affective disorder ( Stegmayer et al., 2015 ). Consistent with the previously described study about emotional processing effects on working memory in older adults, this reported outcome was suggestive of the brain’s failed attempts to suppress pathological amygdalar activation during a verbal working memory task ( Stegmayer et al., 2015 ).

Another affected group with working memory deficits that has been the subject of research interest was children with developmental disorders such as attention deficit/hyperactivity disorder (ADHD), developmental dyscalculia, and reading difficulties ( Rotzer et al., 2009 ; Ashkenazi et al., 2013 ; Wang and Gathercole, 2013 ; Maehler and Schuchardt, 2016 ). For instance, looking into the different working memory subsystems based on Baddeley’s multicomponent working memory model in children with dyslexia and/or ADHD and children with dyscalculia and/or ADHD through a series of tests, it was reported that distinctive working memory deficits by groups could be detected such that phonological loop (e.g., digit span) impairment was observed in the dyslexia group, visuospatial sketchpad (e.g., Corsi block tasks) deficits in the dyscalculia group, while central executive (e.g., complex counting span) deficits in children with ADHD ( Maehler and Schuchardt, 2016 ). Meanwhile, examination of working memory impairment in a delayed match-to-sample visual task that put emphasis on the maintenance phase of working memory by examining the brainwaves of adults with ADHD using electroencephalography (EEG) also revealed a marginally significantly lower alpha band power in the posterior regions as compared to healthy individuals, and such an observation was not significantly improved after working memory training (Cogmed working memory training, CWMT Program) ( Liu et al., 2016 ). The alpha power was considered important in the maintenance of working memory items; and lower working memory accuracy paired with lower alpha band power was indeed observed in the ADHD group ( Liu et al., 2016 ).

Not dismissing the above compiled results, children encountering disabilities in mathematical operations likewise indicated deficits in the working memory domain that were traceable to unusual brain activities at the neurobiological level ( Rotzer et al., 2009 ; Ashkenazi et al., 2013 ). It was speculated that visuospatial working memory plays a vital role when arithmetic problem-solving is involved in order to ensure intact mental representations of the numerical information ( Rotzer et al., 2009 ). Indeed, Ashkenazi et al. (2013) revealed that Block Recall, a variant of the Corsi Block Tapping test and a subtest of the Working Memory Test Battery for Children (WMTB-C) that explored visuospatial sketchpad ability, was significantly predictive of math abilities. In relation to this, studies investigating brain activation patterns and performance of visuospatial working memory task in children with mathematical disabilities identified the intraparietal sulcus (IPS), in conjunction with other regions in the prefrontal and parietal cortices, to have less activation when visuospatial working memory was deemed involved (during an adapted form of Corsi Block Tapping test made suitable for fMRI [ Rotzer et al., 2009 ]); in contrast the control group demonstrated correlations of the IPS in addition to the fronto-parietal cortical activation with the task ( Rotzer et al., 2009 ; Ashkenazi et al., 2013 ). These brain activity variations that translated to differences in overt performances between healthily developing individuals and those with atypical development highlighted the need for intervention and attention for the disadvantaged groups.

Traumatic Brain Injury and Working Memory

Physical injuries impacting the frontal or parietal lobes would reasonably be damaging to one’s working memory. This is supported in studies employing neuropsychological testing to assess cognitive impairments in patients with traumatic brain injury; and poorer cognitive performances especially involving the working memory domains were reported (see Review Articles by Dikmen et al., 2009 ; Dunning et al., 2016 ; Phillips et al., 2017 ). Research on cognitive deficits in traumatic brain injury has been extensive due to the debilitating conditions brought upon an individual daily life after the injury. Traumatic brain injuries (TBI) refer to accidental damage to the brain after being hit by an object or following rapid acceleration or deceleration ( Farrer, 2017 ). These accidents include falls, assaults, or automobile accidents and patients with TBI can be then categorized into three groups; (1) mild TBI with GCS – Glasgow Coma Scale – score of 13–15; (2) moderate TBI with GCS score of 9–12; and (3) severe TBI with GCS score of 3–8 ( Farrer, 2017 ). In a recently published meta-analysis that specifically looked at working memory impairments in patients with moderate to severe TBI, patients displayed reduced cognitive functions in verbal short-term memory in addition to verbal and visuospatial working memory in comparison to control groups ( Dunning et al., 2016 ). It was also understood from the analysis that the time lapse since injury and age of injury were deciding factors that influenced these cognitive deficits in which longer time post-injury or older age during injury were associated with greater cognitive decline ( Dunning et al., 2016 ).

Nonetheless, it is to be noted that such findings relating to age of injury could not be generalized to the child population since results from the pediatric TBI cases showed that damage could negatively impact developmental skills that could indicate a greater lag in cognitive competency as the child’s frontal lobe had yet to mature ( Anderson and Catroppa, 2007 ; Mandalis et al., 2007 ; Nadebaum et al., 2007 ; Gorman et al., 2012 ). These studies all reported working memory impairment of different domains such as attentional control, executive functions, or verbal and visuospatial working memory in the TBI group, especially for children with severe TBI ( Mandalis et al., 2007 ; Nadebaum et al., 2007 ; Gorman et al., 2012 ). Investigation of whether working memory deficits are domain-specific or -general or involve one or more mechanisms, has yielded inconsistent results. For example, Perlstein et al. (2004) found that working memory was impaired in the TBI group only when complex manipulation such as sequential coding of information is required and not accounted for by processing speed or maintenance of information, but two teams of researchers ( Perbal et al., 2003 ; Gorman et al., 2012 ) suggested otherwise. From their study on timing judgments, Perbal et al. (2003) concluded that deficits were not related to time estimation but more on generalized attentional control, working memory and processing speed problems; while Gorman et al. (2012) also attributed the lack of attentional focus to impairments observed during the working memory task. In fact, in a later study by Gorman et al. (2016) , it was shown that processing speed mediated TBI effects on working memory even though the mediation was partial. On the other hand, Vallat-Azouvi et al. (2007) reported impairments in the working memory updating domain that came with high executive demands for TBI patients. Also, Mandalis et al. (2007) similarly highlighted potential problems with attention and taxing cognitive demands in the TBI group.

From the neuroscientific perspective, hyper-activation or -connectivity in the working memory circuitry was reported in TBI patients in comparison with healthy controls when both groups engaged in working memory tasks, suggesting that the brain attempted to compensate for or re-establish lost connections upon the injury ( Dobryakova et al., 2015 ; Hsu et al., 2015 ; Wylie et al., 2015 ). For a start, it was observed that participants with mild TBI displayed increased activation in the right prefrontal cortex during a working memory task when comparing to controls ( Wylie et al., 2015 ). Interestingly, this activation pattern only occurred in patients who did not experience a complete recovery 1 week after the injury ( Wylie et al., 2015 ). Besides, low activation in the DMN was observed in mild TBI patients without cognitive recovery, and such results seemed to be useful in predicting recovery in patients in which the patients did not recover when hypoactivation (low activation) was reported, and vice versa ( Wylie et al., 2015 ). This might be suggestive of the potential of cognitive recovery simply by looking at the intensity of brain activation of the DMN, for an increase in activation of the DMN seemed to be superseded before cognitive recovery was present ( Wylie et al., 2015 ).

In fact, several studies lent support to the speculation mentioned above as hyperactivation or hypoactivation in comparison with healthy participants was similarly identified. When sex differences were being examined in working memory functional activity in mild TBI patients, hyperactivation was reported in male patients when comparing to the male control group, suggesting that the hyperactivation pattern might be the brain’s attempt at recovering impaired functions; even though hypoactivation was shown in female patients as compared to the female control group ( Hsu et al., 2015 ). The researchers from the study further explained that such hyperactivation after the trauma acted as a neural compensatory mechanism so that task performance could be maintained while hypoactivation with a poorer performance could have been the result of a more severe injury ( Hsu et al., 2015 ). Therefore, the decrease in activation in female patients, in addition to the observed worse performance, was speculated to be due to a more serious injury sustained by the female patients group ( Hsu et al., 2015 ).

In addition, investigation of the effective connectivity of moderate and severe TBI participants during a working memory task revealed that the VMPFC influenced the ACC in these TBI participants when the opposite was observed in healthy subjects ( Dobryakova et al., 2015 ). Moreover, increased inter-hemispheric transfer due to an increased number of connections between the left and right hemispheres (hyper-connectivity) without clear directionality of information flow (redundant connectivity) was also reported in the TBI participants ( Dobryakova et al., 2015 ). This study was suggestive of location-specific changes in the neural network connectivity following TBI depending on the cognitive functions at work, other than providing another support to the neural compensatory hypothesis due to the observed hyper-connectivity ( Dobryakova et al., 2015 ).

Nevertheless, inconsistent findings should not be neglected. In a study that also focused on brain connectivity analysis among patients with mild TBI by Hillary et al. (2011) , elevated task-related connectivity in the right hemisphere, in particular the prefrontal cortex, was consistently demonstrated during a working memory task while the control group showed greater left hemispheric activation. This further supported the right lateralization of the brain to reallocate cognitive resources of TBI patients post-injury. Meanwhile, the study did not manage to obtain the expected outcome in terms of greater clustering of whole-brain connections in TBI participants as hypothesized ( Hillary et al., 2011 ). That said, no significant loss or gain of connections due to the injury could be concluded from the study, as opposed to the hyper- or hypoactivation or hyper-connectivity frequently highlighted in other similar researches ( Hillary et al., 2011 ). Furthermore, a study by Chen et al. (2012) also failed to establish the same results of increased brain activation. Instead, with every increase of the working memory load, increase in brain activation, as expected to occur and as demonstrated in the control group, was unable to be detected in the TBI group ( Chen et al., 2012 ).

Taken all the insightful studies together, another aspect not to be neglected is the neuroimaging techniques employed in contributing to the literature on TBI. Modalities other than fMRI, which focuses on localization of brain activities, show other sides of the story of working memory impairments in TBI to offer a more holistic understanding. Studies adopting electroencephalography (EEG) or diffusor tensor imaging (DTI) reported atypical brainwaves coherence or white matter integrity in patients with TBI ( Treble et al., 2013 ; Ellis et al., 2016 ; Bailey et al., 2017 ; Owens et al., 2017 ). Investigating the supero-lateral medial forebrain bundle (MFB) that innervates and consequently terminates at the prefrontal cortex, microstructural white matter damage at the said area was indicated in participants with moderate to severe TBI by comparing its integrity with the control group ( Owens et al., 2017 ). Such observation was backed up by evidence showing that the patients performed more poorly on attention-loaded cognitive tasks of factors relating to slow processing speed than the healthy participants, although a direct association between MFB and impaired attentional system was not found ( Owens et al., 2017 ).

Correspondingly, DTI study of the corpus callosum (CC), which described to hold a vital role in connecting and coordinating both hemispheres to ensure competent cognitive functions, also found compromised microstructure of the CC with low fractional anisotropy and high mean diffusivity, both of which are indications of reduced white matter integrity ( Treble et al., 2013 ). This reported observation was also found to be predictive of poorer verbal or visuospatial working memory performance in callosal subregions connecting the parietal and temporal cortices ( Treble et al., 2013 ). Adding on to these results, using EEG to examine the functional consequences of CC damage revealed that interhemispheric transfer time (IHTT) of the CC was slower in the TBI group than the control group, suggesting an inefficient communication between the two hemispheres ( Ellis et al., 2016 ). In addition, the TBI group with slow IHTT as well exhibited poorer neurocognitive functioning including working memory than the healthy controls ( Ellis et al., 2016 ).

Furthermore, comparing the working memory between TBI, MDD, TBI-MDD, and healthy participants discovered that groups with MDD and TBI-MDD performed poorer on the Sternberg working memory task but functional connectivity on the other hand, showed that increased inter-hemispheric working memory gamma connectivity was observed in the TBI and TBI-MDD groups ( Bailey et al., 2017 ). Speculation provided for the findings of such neuronal state that was not reflected in the explicit working memory performance was that the deficits might not be detected or tested by the utilized Sternberg task ( Bailey et al., 2017 ). Another explanation attempting to answer the increase in gamma connectivity in these groups was the involvement of the neural compensatory mechanism after TBI to improve performance ( Bailey et al., 2017 ). Nevertheless, such outcome implies that behavioral performances or neuropsychological outcomes might not always be reflective of the functional changes happening in the brain.

Yet, bearing in mind that TBI consequences can be vast and crippling, cognitive improvement or recovery, though complicated due to the injury severity-dependent nature, is not impossible (see Review Article by Anderson and Catroppa, 2007 ; Nadebaum et al., 2007 ; Dikmen et al., 2009 ; Chen et al., 2012 ). As reported by Wylie et al. (2015) , cognitive improvement together with functional changes in the brain could be detected in individuals with mild TBI. Increased activation in the brain during 6-week follow-up was also observed in the mild TBI participants, implicating the regaining of connections in the brain ( Chen et al., 2012 ). Administration of certain cognitively enhancing drugs such as methylphenidate was reported to be helpful in improving working memory performance too ( Manktelow et al., 2017 ). Methylphenidate as a dopamine reuptake inhibitor was found to have modulated the neural activity in the left cerebellum which subsequently correlated with improved working memory performance ( Manktelow et al., 2017 ). A simplified summary of recent studies on working memory and TBI is tabulated in Table 4 .

TABLE 4. Working memory (WM) studies in the TBI group.

General Discussion and Future Direction

In practice, all of the aforementioned studies contribute to the working memory puzzle by addressing the topic from different perspectives and employing various methodologies to study it. Several theoretical models of working memory that conceptualized different working memory mechanisms or domains (such as focus of attention, inhibitory controls, maintenance and manipulation of information, updating and integration of information, capacity limits, evaluative and executive controls, and episodic buffer) have been proposed. Coupled with the working memory tasks of various means that cover a broad range (such as Sternberg task, n-back task, Corsi block-tapping test, Wechsler’s Memory Scale [WMS], and working memory subtests in the Wechsler Adult Intelligence Scale [WAIS] – Digit Span, Letter Number Sequencing), it has been difficult, if not highly improbable, for working memory studies to reach an agreement upon a consistent study protocol that is acceptable for generalization of results due to the constraints bound by the nature of the study. Various data acquisition and neuroimaging techniques that come with inconsistent validity such as paper-and-pen neuropsychological measures, fMRI, EEG, DTI, and functional near-infrared spectroscopy (fNIRS), or even animal studies can also be added to the list. This poses further challenges to quantitatively measure working memory as only a single entity. For example, when studying the neural patterns of working memory based on Cowan’s processes-embedded model using fMRI, one has to ensure that the working memory task selected is fMRI-compatible, and demands executive control of attention directed at activated long-term memory (domain-specific). That said, on the one hand, there are tasks that rely heavily on the information maintenance such as the Sternberg task; on the other hand, there are also tasks that look into the information manipulation updating such as the n-back or arithmetic task. Meanwhile, the digit span task in WAIS investigates working memory capacity, although it can be argued that it also encompasses the domain on information maintenance and updating-. Another consideration involves the different natures (verbal/phonological and visuospatial) of the working memory tasks as verbal or visuospatial information is believed to engage differing sensory mechanisms that might influence comparison of working memory performance between tasks of different nature ( Baddeley and Hitch, 1974 ; Cowan, 1999 ). For instance, though both are n-back tasks that includes the same working memory domains, the auditory n-back differs than the visual n-back as the information is presented in different forms. This feature is especially crucial with regards to the study populations as it differentiates between verbal and visuospatial working memory competence within individuals, which are assumed to be domain-specific as demonstrated by vast studies (such as Nadler and Archibald, 2014 ; Pham and Hasson, 2014 ; Nakagawa et al., 2016 ). These test variations undeniably present further difficulties in selecting an appropriate task. Nevertheless, the adoption of different modalities yielded diverging outcomes and knowledge such as behavioral performances, functional segregation and integration in the brain, white matter integrity, brainwave coherence, and oxy- and deoxyhaemoglobin concentrations that are undeniably useful in application to different fields of study.

In theory, the neural efficiency hypothesis explains that increased efficiency of the neural processes recruit fewer cerebral resources in addition to displaying lower activation in the involved neural network ( Vartanian et al., 2013 ; Rodriguez Merzagora et al., 2014 ). This is in contrast with the neural compensatory hypothesis in which it attempted to understand diminished activation that is generally reported in participants with TBI ( Hillary et al., 2011 ; Dobryakova et al., 2015 ; Hsu et al., 2015 ; Wylie et al., 2015 ; Bailey et al., 2017 ). In the diseased brain, low activation has often been associated with impaired cognitive function ( Chen et al., 2012 ; Dobryakova et al., 2015 ; Wylie et al., 2015 ). Opportunely, the CRUNCH model proposed within the field of aging might be translated and integrated the two hypotheses here as it suitably resolved the disparity of cerebral hypo- and hyper-activation observed in weaker, less efficient brains as compared to healthy, adept brains ( Reuter-Lorenz and Park, 2010 ; Schneider-Garces et al., 2010 ). Moreover, other factors such as the relationship between fluid intelligence and working memory might complicate the current understanding of working memory as a single, isolated construct since working memory is often implied in measurements of the intelligence quotient ( Cowan, 2008 ; Vartanian et al., 2013 ). Indeed, the process overlap theory of intelligence proposed by Kovacs and Conway (2016) in which the constructs of intelligence were heavily scrutinized (such as general intelligence factors, g and its smaller counterparts, fluid intelligence or reasoning, crystallized intelligence, perceptual speed, and visual-spatial ability), and fittingly connected working memory capacity with fluid reasoning. Cognitive tests such as Raven’s Progressive Matrices or other similar intelligence tests that demand complex cognition and were reported in the paper had been found to correlate strongly with tests of working memory ( Kovacs and Conway, 2016 ). Furthermore, in accordance with such views, in the same paper, neuroimaging studies found intelligence tests also activated the same fronto-parietal network observed in working memory ( Kovacs and Conway, 2016 ).

On the other hand, even though the roles of the prefrontal cortex in working memory have been widely established, region specificity and localization in the prefrontal cortex in relation to the different working memory domains such as manipulation or delayed retention of information remain at the premature stage (see Review Article by D’Esposito and Postle, 2015 ). It has been postulated that the neural mechanisms involved in working memory are of high-dimensionality and could not always be directly captured and investigated using neurophysiological techniques such as fMRI, EEG, or patch clamp recordings even when comparing with lesion data ( D’Esposito and Postle, 2015 ). According to D’Esposito and Postle (2015) , human fMRI studies have demonstrated that a rostral-caudal functional gradient related to level of abstraction required of working memory along the frontal cortex (in which different regions in the prefrontal cortex [from rostral to caudal] might be associated with different abstraction levels) might exist. Other functional gradients relating to different aspects of working memory were similarly unraveled ( D’Esposito and Postle, 2015 ). These proposed mechanisms with different empirical evidence point to the fact that conclusive understanding regarding working memory could not yet be achieved before the inconsistent views are reconciled.

Not surprisingly, with so many aspects of working memory yet to be understood and its growing complexity, the cognitive neuroscience basis of working memory requires constant research before an exhaustive account can be gathered. From the psychological conceptualization of working memory as attempted in the multicomponent working memory model ( Baddeley and Hitch, 1974 ), to the neural representations of working memory in the brain, especially in the frontal regions ( D’Esposito and Postle, 2015 ), one important implication derives from the present review of the literatures is that working memory as a psychological construct or a neuroscientific mechanism cannot be investigated as an isolated event. The need for psychology and neuroscience to interact with each other in an active feedback cycle exists in which this cognitive system called working memory can be dissected at the biological level and refined both empirically, and theoretically.

In summary, the present article offers an account of working memory from the psychological and neuroscientific perspectives, in which theoretical models of working memory are presented, and neural patterns and brain regions engaging in working memory are discussed among healthy and diseased brains. It is believed that working memory lays the foundation for many other cognitive controls in humans, and decoding the working memory mechanisms would be the first step in facilitating understanding toward other aspects of human cognition such as perceptual or emotional processing. Subsequently, the interactions between working memory and other cognitive systems could reasonably be examined.

Author Contributions

WC wrote the manuscript with critical feedback and consultation from AAH. WC and AAH contributed to the final version of the manuscript. JA supervised the process and proofread the manuscript.

This work was supported by the Transdisciplinary Research Grant Scheme (TRGS) 203/CNEURO/6768003 and the USAINS Research Grant 2016.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer EB and handling Editor declared their shared affiliation.

Andersen, R. A., and Cui, H. (2009). Review intention, action planning, and decision making in parietal-frontal circuits. Neuron 63, 568–583. doi: 10.1016/j.neuron.2009.08.028

PubMed Abstract | CrossRef Full Text | Google Scholar

Anderson, V., and Catroppa, C. (2007). Memory outcome at 5 years post-childhood traumatic brain injury. Brain Inj. 21, 1399–1409. doi: 10.1080/02699050701785070

Arendash, G. W., Schleif, W., Rezai-Zadeh, K., Jackson, E. K., Zacharia, L. C., Cracchiolo, J. R., et al. (2006). Caffeine protects Alzheimer’s mice against cognitive impairment and reduces brain β-amyloid production. Neuroscience 142, 941–952. doi: 10.1016/j.neuroscience.2006.07.021

Ashkenazi, S., Rosenberg-lee, M., Metcalfe, A. W. S., Swigart, A. G., and Menon, V. (2013). Neuropsychologia visuo – spatial working memory is an important source of domain-general vulnerability in the development of arithmetic cognition. Neuropsychologia 51, 2305–2317. doi: 10.1016/j.neuropsychologia.2013.06.031

Baars, B. J., and Franklin, S. (2003). How conscious experience and working memory interact. Trends Cogn. Sci. 7, 166–172. doi: 10.1016/S1364-6613(03)00056-1

CrossRef Full Text | Google Scholar

Baddeley, A. (1996). Exploring the central executive. Q. J. Exp. Psychol. A 49, 5–28. doi: 10.1080/713755608

Baddeley, A. (2010). Working memory. Curr. Biol. 20, R136–R140. doi: 10.1016/j.cub.2009.12.014

Baddeley, A. (2012). Working memory: theories, models, and controversies. Annu. Rev. Psychol. 63, 1–29. doi: 10.1146/annurev-psych-120710-100422

Baddeley, A., and Hitch, G. (1974). Working memory. Psychol. Learn. Motiv. 8, 47–89. doi: 10.1016/j.cub.2009.12.014

Baddeley, A. D. (2000a). The episodic buffer : a new component of working memory? Trends Cogn. Sci. 4, 417–423. doi: 10.1016/S1364-6613(00)01538-2

Baddeley, A. D. (2000b). Short-Term and Working Memory. The Oxford Handbook of Memory. Oxford: Oxford University Press.

Google Scholar

Bailey, N. W., Rogasch, N. C., Hoy, K. E., Maller, J. J., Segrave, R. A., Sullivan, C. M., et al. (2017). Increased gamma connectivity during working memory retention following traumatic brain injury. Brain Inj. 31, 379–389. doi: 10.1080/02699052.2016.1239273

Balderston, N. L., Vytal, K. E., O’Connell, K., Torrisi, S., Letkiewicz, A., Ernst, M., et al. (2017). Anxiety patients show reduced working memory related dlPFC activation during safety and threat. Depress. Anxiety 34, 25–36. doi: 10.1002/da.22518

Barrouillet, P., Bernardin, S., and Camos, V. (2004). Time constraints and resource sharing in adults’ working memory spans. J. Exp. Psychol. Gen. 133, 83–100. doi: 10.1037/0096-3445.133.1.83

Barrouillet, P., and Camos, V. (2007). “The time-based resource-sharing model of working memory,” in The Cognitive Neuroscience of Working Memory , ed. N. Osaka (Oxford: Oxford University Press), 59–80. doi: 10.1093/acprof:oso/9780198570394.003.0004

Barrouillet, P., Gavens, N., Vergauwe, E., Gaillard, V., and Camos, V. (2009). Working memory span development: a time-based resource-sharing model account. Dev. Psychol. 45, 477–490. doi: 10.1037/a0014615

Bolkan, S. S., Stujenske, J. M., Parnaudeau, S., Spellman, T. J., Rauffenbart, C., Abbas, A. I., et al. (2017). Thalamic projections sustain prefrontal activity during working memory maintenance. Nat. Neurosci. 20, 987–996. doi: 10.1038/nn.4568

Borella, E., Carretti, B., Sciore, R., Capotosto, E., Taconnat, L., Cornoldi, C., et al. (2017). Training working memory in older adults: is there an advantage of using strategies? Psychol. Aging 32, 178–191. doi: 10.1037/pag0000155

Chein, J. M., Moore, A. B., and Conway, A. R. A. (2011). NeuroImage domain-general mechanisms of complex working memory span. Neuroimage 54, 550–559. doi: 10.1016/j.neuroimage.2010.07.067

Chen, C. J., Wu, C. H., Liao, Y. P., Hsu, H. L., Tseng, Y. C., Liu, H. L., et al. (2012). Working memory in patients with mild traumatic brain injury: functional MR imaging analysis. Radiology 264, 844–851. doi: 10.1148/radiol.12112154

Cowan, N. (1999). “An embedded-processes model of working memory,” in Models of Working Memory: Mechanisms of Active Maintenance and Executive Control , eds A. Miyake and P. Shah (Cambridge: Cambridge University Press). doi: 10.1017/S0140525X01003922

Cowan, N. (2005). Working memory capacity. Exp. Psychol. 54, 245–246. doi: 10.1027/1618-3169.54.3.245

Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Prog. Brain Res. 169, 323–338. doi: 10.1016/S0079-6123(07)00020-9

Cowan, N. (2010). The magical mystery four. Curr. Dir. Psychol. Sci. 19, 51–57. doi: 10.1177/0963721409359277

Dahlin, E., Nyberg, L., Bäckman, L., and Neely, A. S. (2008). Plasticity of executive functioning in young and older adults: immediate training gains, transfer, and long-term maintenance. Psychol. Aging 23, 720–730. doi: 10.1037/a0014296

Daneman, M., and Carpenter, P. A. (1980). Individual differences in working memory and reading. J. Verbal Learn. Verbal Behav. 19, 450–466. doi: 10.1016/S0022-5371(80)90312-6

D’Esposito, M., and Postle, B. R. (2015). The cognitive neuroscience of working memory. Annu. Rev. Psychol. 66, 115–142. doi: 10.1146/annurev-psych-010814-015031

Dikmen, S. S., Corrigan, J. D., Levin, H. S., Machamer, J., Stiers, W., and Weisskopf, M. G. (2009). Cognitive outcome following traumatic brain injury. J. Head Trauma Rehabil. 24, 430–438. doi: 10.1097/HTR.0b013e3181c133e9

Dima, D., Jogia, J., and Frangou, S. (2014). Dynamic causal modeling of load-dependent modulation of effective connectivity within the verbal working memory network. Hum. Brain Mapp. 35, 3025–3035. doi: 10.1002/hbm.22382

Dobryakova, E., Boukrina, O., and Wylie, G. R. (2015). Investigation of information flow during a novel working memory task in individuals with traumatic brain injury. Brain Connect. 5, 433–441. doi: 10.1089/brain.2014.0283

Duff, S. J., and Hampson, E. (2000). A beneficial effect of estrogen on working memory in postmenopausal women taking hormone replacement therapy. Horm. Behav. 38, 262–276. doi: 10.1006/hbeh.2000.1625

Dunning, D. L., Westgate, B., and Adlam, A.-L. R. (2016). A meta-analysis of working memory impairments in survivors of moderate-to-severe traumatic brain injury. Neuropsychology 30, 811–819. doi: 10.1037/neu0000285

Ellis, M. U., DeBoard Marion, S., McArthur, D. L., Babikian, T., Giza, C., Kernan, C. L., et al. (2016). The UCLA study of children with moderate-to-severe traumatic brain injury: event-related potential measure of interhemispheric transfer time. J. Neurotrauma 33, 990–996. doi: 10.1089/neu.2015.4023

Engle, R. W. (2002). Working memory capacity as executive attention. Curr. Dir. Psychol. Sci. 11, 19–23. doi: 10.1111/1467-8721.00160

Engle, R. W., and Kane, M. J. (2004). “Executive attention, working memory capacity, and a two-factor theory of cognitive control,” in The Psychology of Learning and Motivation: Advances in Research and Theory , ed. B. H. Ross (New York, NY: Elsevier), 145–199. doi: 10.1016/S0079-7421(03)44005-X

Farrer, T. J. (2017). Encyclopedia of Geropsychology , ed. N. A. Pachana. Singapore: Springer. doi: 10.1007/978-981-287-080-3

CrossRef Full Text

Fiebig, F., and Lansner, A. (2017). A spiking working memory model based on hebbian short-term potentiation. J. Neurosci. 37, 83–96. doi: 10.1523/JNEUROSCI.1989-16.2016

Friston, K., Moran, R., and Seth, A. K. (2013). Analysing connectivity with granger causality and dynamic causal modelling. Curr. Opin. Neurobiol. 23, 172–178. doi: 10.1016/j.conb.2012.11.010

Gorman, S., Barnes, M. A., Swank, P. R., Prasad, M., Cox, C. S., and Ewing-Cobbs, L. (2016). Does processing speed mediate the effect of pediatric traumatic brain injury on working memory? Neuropsychology 30, 263–273. doi: 10.1037/neu0000214

Gorman, S., Barnes, M. A., Swank, P. R., Prasad, M., and Ewing-Cobbs, L. (2012). The effects of pediatric traumatic brain injury on verbal and visual-spatial working memory. J. Int. Neuropsychol. Soc. 18, 29–38. doi: 10.1017/S1355617711001251

Gottwald, B., Wilde, B., Mihajlovic, Z., and Mehdorn, H. M. (2004). Evidence for distinct cognitive deficits after focal cerebellar lesions. J. Neurol. Neurosurg. Psychiatry 75, 1524–1531. doi: 10.1136/jnnp.2003.018093

Grot, S., Légaré, V. P., Lipp, O., Soulières, I., Dolcos, F., and Luck, D. (2017). Abnormal prefrontal and parietal activity linked to deficient active binding in working memory in schizophrenia. Schizophr. Res. 188, 68–74. doi: 10.1016/j.schres.2017.01.021

Guye, S., and von Bastian, C. C. (2017). Working memory training in older adults: bayesian evidence supporting the absence of transfer. Psychol. Aging 32, 732–746. doi: 10.1037/pag0000206

Haller, S., Montandon, M.-L., Rodriguez, C., Moser, D., Toma, S., Hofmeister, J., et al. (2017). Caffeine impact on working memory-related network activation patterns in early stages of cognitive decline. Neuroradiology 59, 387–395. doi: 10.1007/s00234-017-1803-5

Haller, S., Rodriguez, C., Moser, D., Toma, S., Hofmeister, J., Sinanaj, I., et al. (2013). Acute caffeine administration impact on working memory-related brain activation and functional connectivity in the elderly: a BOLD and perfusion MRI study. Neuroscience 250, 364–371. doi: 10.1016/j.neuroscience.2013.07.021

Hedden, T., and Gabrieli, J. D. E. (2004). Insights into the ageing mind: a view from cognitive neuroscience. Nat. Rev. Neurosci. 5, 87–96. doi: 10.1038/nrn1323

Heinzel, S., Rimpel, J., Stelzel, C., and Rapp, M. A. (2017). Transfer effects to a multimodal dual-task after working memory training and associated neural correlates in older adults – a pilot study. Front. Hum. Neurosci. 11:85. doi: 10.3389/fnhum.2017.00085

Hillary, F. G., Medaglia, J. D., Gates, K., Molenaar, P. C., Slocomb, J., Peechatka, A., et al. (2011). Examining working memory task acquisition in a disrupted neural network. Brain 134, 1555–1570. doi: 10.1093/brain/awr043

Hsu, H.-L., Chen, D. Y.-T., Tseng, Y.-C., Kuo, Y.-S., Huang, Y.-L., Chiu, W.-T., et al. (2015). Sex differences in working memory after mild traumatic brain injury: a functional MR imaging study. Radiology 276, 828–835. doi: 10.1148/radiol.2015142549

Humphreys, M. S., Bain, J. D., and Pike, R. (1989). Different ways to cue a coherent memory system: a theory for episodic, semantic, and procedural tasks. Psychol. Rev. 96, 208–233. doi: 10.1037/0033-295X.96.2.208

Janowsky, J. S., Chavez, B., and Orwoll, E. (2000). Sex steroids modify working memory. J. Cogn. Neurosci. 12, 407–414. doi: 10.1162/089892900562228

Jimura, K., Chushak, M. S., Westbrook, A., and Braver, T. S. (2017). Intertemporal decision-making involves prefrontal control mechanisms associated with working memory. Cereb. Cortex doi: 10.1093/cercor/bhx015 [Epub ahead of print].

Joseph, J. E., Swearingen, J. E., Corbly, C. R., Curry, T. E., and Kelly, T. H. (2012). Influence of estradiol on functional brain organization for working memory. Neuroimage 59, 2923–2931. doi: 10.1016/j.neuroimage.2011.09.067

Karbach, J., and Verhaeghen, P. (2014). Making working memory work: a meta-analysis of executive control and working memory training in younger and older adults. Psychol. Sci. 25, 2027–2037. doi: 10.1177/0956797614548725

Kim, C., Kroger, J. K., Calhoun, V. D., and Clark, V. P. (2015). The role of the frontopolar cortex in manipulation of integrated information in working memory. Neurosci. Lett. 595, 25–29. doi: 10.1016/j.neulet.2015.03.044

Klaassen, E. B., De Groot, R. H. M., Evers, E. A. T., Snel, J., Veerman, E. C. I., Ligtenberg, A. J. M., et al. (2013). The effect of caffeine on working memory load-related brain activation in middle-aged males. Neuropharmacology 64, 160–167. doi: 10.1016/j.neuropharm.2012.06.026

Kovacs, K., and Conway, A. R. A. (2016). Process overlap theory: a unified account of the general factor of intelligence. Psychol. Inq. 27, 151–177. doi: 10.1080/1047840X.2016.1153946

Le, T. M., Borghi, J. A., Kujawa, A. J., Klein, D. N., and Leung, H.-C. (2017). Alterations in visual cortical activation and connectivity with prefrontal cortex during working memory updating in major depressive disorder. Neuroimage 14, 43–53. doi: 10.1016/j.nicl.2017.01.004

Liu, Z.-X., Glizer, D., Tannock, R., and Woltering, S. (2016). EEG alpha power during maintenance of information in working memory in adults with ADHD and its plasticity due to working memory training: a randomized controlled trial. Clin. Neurophysiol. 127, 1307–1320. doi: 10.1016/j.clinph.2015.10.032

Ma, L., Steinberg, J. L., Hasan, K. M., Narayana, P. A., Kramer, L. A., and Moeller, F. G. (2012). Working memory load modulation of parieto-frontal connections: evidence from dynamic causal modeling. Hum. Brain Mapp. 33, 1850–1867. doi: 10.1002/hbm.21329

Maehler, C., and Schuchardt, K. (2016). Working memory in children with specific learning disorders and/or attention deficits. Learn. Individ. Differ. 49, 341–347. doi: 10.1016/j.lindif.2016.05.007

Mandalis, A., Kinsella, G., Ong, B., and Anderson, V. (2007). Working memory and new learning following pediatric traumatic brain injury. Dev. Neuropsychol. 32, 683–701. doi: 10.1080/87565640701376045

Manktelow, A. E., Menon, D. K., Sahakian, B. J., and Stamatakis, E. A. (2017). Working memory after traumatic brain injury: the neural basis of improved performance with methylphenidate. Front. Behav. Neurosci. 11:58. doi: 10.3389/fnbeh.2017.00058

Miller, G. A., Galanter, E., and Pribram, K. H. (1960). Plans and the Structure of Behavior. New York, NY: Henry Holt and Company. doi: 10.1037/10039-000

Miyake, A., and Shah, P. (eds). (1999). Models of Working Memory: Mechanisms of Active Maintenance and Executive Control. New York, NY: Cambridge University Press. doi: 10.1017/CBO9781139174909

Mongillo, G., Barak, O., and Tsodyks, M. (2008). Synaptic theory of working memory. Science 319, 1543–1546. doi: 10.1126/science.1150769

Moore, A. B., Li, Z., Tyner, C. E., Hu, X., and Crosson, B. (2013). Bilateral basal ganglia activity in verbal working memory. Brain Lang. 125, 316–323. doi: 10.1016/j.bandl.2012.05.003

Murty, V. P., Sambataro, F., Radulescu, E., Altamura, M., Iudicello, J., Zoltick, B., et al. (2011). Selective updating of working memory content modulates meso-cortico-striatal activity. Neuroimage 57, 1264–1272. doi: 10.1016/j.neuroimage.2011.05.006

Nadebaum, C., Anderson, V., and Catroppa, C. (2007). Executive function outcomes following traumatic brain injury in young children: a five year follow-up. Dev. Neuropsychol. 32, 703–728. doi: 10.1080/87565640701376086

Nadler, R. T., and Archibald, L. M. D. (2014). The assessment of verbal and visuospatial working memory with school age canadian children. Can. J. Speech Lang. Pathol. Audiol. 38, 262–279.

Nakagawa, S., Takeuchi, H., Taki, Y., Nouchi, R., Sekiguchi, A., Kotozaki, Y., et al. (2016). Sex-related differences in the effects of sleep habits on verbal and visuospatial working memory. Front. Psychol. 7:1128. doi: 10.3389/fpsyg.2016.01128

Nissim, N. R., O’Shea, A. M., Bryant, V., Porges, E. C., Cohen, R., and Woods, A. J. (2017). Frontal structural neural correlates of working memory performance in older adults. Front. Aging Neurosci. 8:328. doi: 10.3389/fnagi.2016.00328

Oren, N., Ash, E. L., Tarrasch, R., Hendler, T., Giladi, N., and Shapira-Lichter, I. (2017). Neural patterns underlying the effect of negative distractors on working memory in older adults. Neurobiol. Aging 53, 93–102. doi: 10.1016/j.neurobiolaging.2017.01.020

Osaka, M., Osaka, N., Kondo, H., Morishita, M., Fukuyama, H., Aso, T., et al. (2003). The neural basis of individual differences in working memory capacity: an fMRI study. Neuroimage 18, 789–797. doi: 10.1016/S1053-8119(02)00032-0

Owen, A. M., McMillan, K. M., Laird, A. R., and Bullmore, E. (2005). N-back working memory paradigm: a meta-analysis of normative functional neuroimaging studies. Hum. Brain Mapp. 25, 46–59. doi: 10.1002/hbm.20131

Owens, J. A., Spitz, G., Ponsford, J. L., Dymowski, A. R., Ferris, N., and Willmott, C. (2017). White matter integrity of the medial forebrain bundle and attention and working memory deficits following traumatic brain injury. Brain Behav. 7:e00608. doi: 10.1002/brb3.608

Perbal, S., Couillet, J., Azouvi, P., and Pouthas, V. (2003). Relationships between time estimation, memory, attention, and processing speed in patients with severe traumatic brain injury. Neuropsychologia 41, 1599–1610. doi: 10.1016/S0028-3932(03)00110-6

Perlstein, W. M., Cole, M. A., Demery, J. A., Seignourel, P. J., Dixit, N. K., Larson, M. J., et al. (2004). Parametric manipulation of working memory load in traumatic brain injury: behavioral and neural correlates. J. Int. Neuropsychol. Soc. 10, 724–741. doi: 10.1017/S1355617704105110

Pham, A. V., and Hasson, R. M. (2014). Verbal and visuospatial working memory as predictors of children’s reading ability. Arch. Clin. Neuropsychol. 29, 467–477. doi: 10.1093/arclin/acu024

Phillips, N. L., Parry, L., Mandalis, A., and Lah, S. (2017). Working memory outcomes following traumatic brain injury in children: a systematic review with meta-analysis. Child Neuropsychol. 23, 26–66. doi: 10.1080/09297049.2015.1085500

Rees, K., Allen, D., and Lader, M. (1999). The influences of age and caffeine on psychomotor and cognitive function. Psychopharmacology 145, 181–188. doi: 10.1007/s002130051047

Reuter-Lorenz, P. A., and Cappell, K. A. (2008). Neurocognitive ageing and the compensation hypothesis. Curr. Dir. Psychol. Sci. 17, 177–182. doi: 10.1111/j.1467-8721.2008.00570.x

Reuter-Lorenz, P. A., and Park, D. C. (2010). Human neuroscience and the aging mind : a new look at old problems. J. Gerontol. Psychol. Sci. 65, 405–415. doi: 10.1093/geronb/gbq035

Rieck, J. R., Rodrigue, K. M., Boylan, M. A., and Kennedy, K. M. (2017). Age-related reduction of BOLD modulation to cognitive difficulty predicts poorer task accuracy and poorer fluid reasoning ability. Neuroimage 147, 262–271. doi: 10.1016/j.neuroimage.2016.12.022

Rodriguez Merzagora, A. C., Izzetoglu, M., Onaral, B., and Schultheis, M. T. (2014). Verbal working memory impairments following traumatic brain injury: an fNIRS investigation. Brain Imaging Behav. 8, 446–459. doi: 10.1007/s11682-013-9258-8

Rose, N. S., LaRocque, J. J., Riggall, A. C., Gosseries, O., Starrett, M. J., Meyering, E. E., et al. (2016). Reactivation of latent working memories with transcranial magnetic stimulation. Science 354, 1136–1139. doi: 10.1126/science.aah7011

Rotzer, S., Loenneker, T., Kucian, K., Martin, E., Klaver, P., and von Aster, M. (2009). Dysfunctional neural network of spatial working memory contributes to developmental dyscalculia. Neuropsychologia 47, 2859–2865. doi: 10.1016/j.neuropsychologia.2009.06.009

Schneider-Garces, N. J., Gordon, B. A., Brumback-Peltz, C. R., Shin, E., Lee, Y., Sutton, B. P., et al. (2010). Span, CRUNCH, and beyond: working memory capacity and the aging brain. J. Cogn. Neurosci. 22, 655–669. doi: 10.1162/jocn.2009.21230

Schöning, S., Engelien, A., Kugel, H., Schäfer, S., Schiffbauer, H., Zwitserlood, P., et al. (2007). Functional anatomy of visuo-spatial working memory during mental rotation is influenced by sex, menstrual cycle, and sex steroid hormones. Neuropsychologia 45, 3203–3214. doi: 10.1016/j.neuropsychologia.2007.06.011

Silvanto, J. (2017). Working memory maintenance: sustained firing or synaptic mechanisms? Trends Cogn. Sci. 21, 152–154. doi: 10.1016/j.tics.2017.01.009

Stegmayer, K., Usher, J., Trost, S., Henseler, I., Tost, H., Rietschel, M., et al. (2015). Disturbed cortico–amygdalar functional connectivity as pathophysiological correlate of working memory deficits in bipolar affective disorder. Eur. Arch. Psychiatry Clin. Neurosci. 265, 303–311. doi: 10.1007/s00406-014-0517-5

Treble, A., Hasan, K. M., Iftikhar, A., Stuebing, K. K., Kramer, L. A., Cox, C. S., et al. (2013). Working memory and corpus callosum microstructural integrity after pediatric traumatic brain injury: a diffusion tensor tractography study. J. Neurotrauma 30, 1609–1619. doi: 10.1089/neu.2013.2934

Vallat-Azouvi, C., Weber, T., Legrand, L., and Azouvi, P. (2007). Working memory after severe traumatic brain injury. J. Int. Neuropsychol. Soc. 13, 770–780. doi: 10.1017/S1355617707070993

Vartanian, O., Jobidon, M.-E., Bouak, F., Nakashima, A., Smith, I., Lam, Q., et al. (2013). Working memory training is associated with lower prefrontal cortex activation in a divergent thinking task. Neuroscience 236, 186–194. doi: 10.1016/j.neuroscience.2012.12.060

Wang, S., and Gathercole, S. E. (2013). Working memory deficits in children with reading difficulties: memory span and dual task coordination. J. Exp. Child Psychol. 115, 188–197. doi: 10.1016/j.jecp.2012.11.015

Wylie, G. R., Freeman, K., Thomas, A., Shpaner, M., OKeefe, M., Watts, R., et al. (2015). Cognitive improvement after mild traumatic brain injury measured with functional neuroimaging during the acute period. PLoS One 10:e0126110. doi: 10.1371/journal.pone.0126110

Ziaei, M., Salami, A., and Persson, J. (2017). Age-related alterations in functional connectivity patterns during working memory encoding of emotional items. Neuropsychologia 94, 1–12. doi: 10.1016/j.neuropsychologia.2016.11.012

Ziemus, B., Baumann, O., Luerding, R., Schlosser, R., Schuierer, G., Bogdahn, U., et al. (2007). Impaired working-memory after cerebellar infarcts paralleled by changes in bold signal of a cortico-cerebellar circuit. Neuropsychologia 45, 2016–2024. doi: 10.1016/j.neuropsychologia.2007.02.012

Zylberberg, J., and Strowbridge, B. W. (2017). Mechanisms of persistent activity in cortical circuits: possible neural substrates for working memory. Annu. Rev. Neurosci. 40, 603–627. doi: 10.1146/annurev-neuro-070815-014006

Keywords : working memory, neuroscience, psychology, cognition, brain, central executive, prefrontal cortex, review

Citation: Chai WJ, Abd Hamid AI and Abdullah JM (2018) Working Memory From the Psychological and Neurosciences Perspectives: A Review. Front. Psychol. 9:401. doi: 10.3389/fpsyg.2018.00401

Received: 24 November 2017; Accepted: 09 March 2018; Published: 27 March 2018.

Reviewed by:

Copyright © 2018 Chai, Abd Hamid and Abdullah. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Aini Ismafairus Abd Hamid, [email protected]

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Loading metrics

Open Access

Peer-reviewed

Research Article

Beta-band neural variability reveals age-related dissociations in human working memory maintenance and deletion

Roles Conceptualization, Data curation, Formal analysis, Software, Visualization, Writing – original draft

Affiliation Department of Psychological and Brain Sciences, Boston University, Boston, Massachusetts, United States of America

Roles Conceptualization, Formal analysis, Writing – review & editing

Roles Data curation, Formal analysis

Affiliation Tufts University, Department of Biology, Medford, Massachusetts, United States of America

Roles Data curation

Roles Formal analysis

Roles Data curation, Software

Roles Conceptualization, Funding acquisition, Supervision, Writing – review & editing

* E-mail: [email protected]

Affiliations Department of Psychological and Brain Sciences, Boston University, Boston, Massachusetts, United States of America, Department of Biomedical Engineering, Boston University, Boston, Massachusetts, United States of America, Center for Systems Neuroscience, Boston University, Boston, Massachusetts, United States of America, Cognitive Neuroimaging Center, Boston University, Boston, Massachusetts, United States of America, Center for Research in Sensory Communication and Emerging Neural Technology, Boston University, Boston, Massachusetts, United States of America

• Wen Wen, 
• Shrey Grover, 
• Douglas Hazel, 
• Peyton Berning, 
• Frederik Baumgardt, 
• Vighnesh Viswanathan, 
• Olivia Tween, 
• Robert M. G. Reinhart

• Published: September 11, 2024
• https://doi.org/10.1371/journal.pbio.3002784
• Reader Comments

Maintaining and removing information in mind are 2 fundamental cognitive processes that decline sharply with age. Using a combination of beta-band neural oscillations, which have been implicated in the regulation of working memory contents, and cross-trial neural variability, an undervalued property of brain dynamics theorized to govern adaptive cognitive processes, we demonstrate an age-related dissociation between distinct working memory functions—information maintenance and post-response deletion. Load-dependent decreases in beta variability during maintenance predicted memory performance of younger, but not older adults. Surprisingly, the post-response phase emerged as the predictive locus of working memory performance for older adults, with post-response beta variability correlated with memory performance of older, but not younger adults. Single-trial analysis identified post-response beta power elevation as a frequency-specific signature indexing memory deletion. Our findings demonstrate the nuanced interplay between age, beta dynamics, and working memory, offering valuable insights into the neural mechanisms of cognitive decline in agreement with the inhibition deficit theory of aging.

Citation: Wen W, Grover S, Hazel D, Berning P, Baumgardt F, Viswanathan V, et al. (2024) Beta-band neural variability reveals age-related dissociations in human working memory maintenance and deletion. PLoS Biol 22(9): e3002784. https://doi.org/10.1371/journal.pbio.3002784

Academic Editor: Frank Tong, Vanderbilt University, UNITED STATES OF AMERICA

Received: December 18, 2023; Accepted: August 2, 2024; Published: September 11, 2024

Copyright: © 2024 Wen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Object stimuli are available at https://bradylab.ucsd.edu/stimuli.html . Scripts and source data are available at Zenodo, DOI 10.5281/zenodo.12735828 , at https://doi.org/10.5281/zenodo.12735828 .

Funding: RMGR is supported by grants from the National Institutes of Health (R01-MH114877; R01-AG063775; R01-AG082645), the International Obsessive-Compulsive Disorder Foundation (IOCDF), the AE Research Foundation, and philanthropy. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Abbreviations: BEM, boundary element method; CRUNCH, Compensation-Related Utilization of Neural Circuits Hypothesis; EEG, electroencephalography; GLMM, generalized linear mixed model; ICA, independent component analysis; ISI, interstimulus interval; LCMV, linearly constrained minimum variance; RT, reaction time; SNR, signal-to-noise ratio

Introduction

Working memory is a basic cognitive function markedly affected by aging [ 1 , 2 ]. Efficient working memory function is facilitated by multiple processes. On the one hand, processes that promote maintenance of information are important [ 3 ]. Emerging research has identified the neural mechanisms contributing to maintenance deficits with age [ 4 ]. On the other hand, processes that remove information when it loses its relevance are equally important [ 5 ]. Failure to remove irrelevant thoughts from mind can obstruct our capacity-limited systems and interfere with the maintenance of relevant information [ 6 , 7 ]. In fact, a leading theory of neurocognitive aging—the inhibition deficit theory—suggests that impairments in the ability to delete information from working memory are what primarily contribute to age-related decline [ 8 ]. Despite the considerable body of work on age-related deletion deficits in distractor inhibition [ 9 ], only limited attention has been given to the deletion of targets after responses, with no reference to the underlying neural mechanisms.

Beyond its well-studied role in sensorimotor control, rhythmic neural activity in the beta band (15 to 25 Hz) has been suggested to regulate the status of working memory contents [ 10 – 12 ]. Dynamics in beta-band activity reflect working memory processing. There is a decrease in beta activity when information needs to be maintained and an increase when information needs to be deleted [ 13 ]. Maintenance-related beta decrease is primarily observed in the prefrontal cortex [ 13 , 14 ]. By contrast, post-response beta increase is observed among task-related networks involving frontal and centroparietal regions [ 10 ], facilitating removal of both memory contents and associated representations such as motor plans after responses. Specifically, neurophysiological evidence from nonhuman primates demonstrated localized post-response beta increase at sites containing memory information during the time course of working memory clear-out [ 13 ]. Whether such dynamics can be observed in human electrophysiology and how these neural dynamics change with age is unknown.

We examined the neural mechanisms underlying age-related decline in multiple working memory phases. To accommodate the increased interindividual variability in cognitive aging [ 2 , 15 ], we were further interested in studying neural metrics that are capable of characterizing individual differences in both younger and older adults. Neural variability is an understudied property of brain dynamics, which is increasingly recognized as a sensitive index capable of tracking intra- and interindividual brain–behavior relationships [ 16 – 18 ]. It reflects the joint influence of sensory input, arousal state, attention, and high-order demand variations on brain functions [ 18 ]. In particular, behavioral relevance of cross-trial variability has been reported in multiple research fields, with lower variability associated with superior perception [ 19 ], more internally guided decision-making [ 20 ], and less social conformity behavior [ 21 ]. Thus, we leveraged cross-trial neural variability to examine the beta-band oscillatory dynamics during maintenance and after response, with a particular focus on age-related differences.

Pronounced age-related working memory deficits with increasing set size

Twenty younger (22.1 ± 2.5 years) and 21 older (70.6 ± 4.8 years) adults performed a delayed match-to-sample task involving 1, 2, or 4 sequentially presented real-world objects during concurrent electroencephalography (EEG) ( Fig 1A ). Participants were instructed to indicate using a corresponding button press whether a subsequently presented probe item was identical to any one of their memory representations. Feedback was presented 0.5 s after response. Behavioral performance accuracy was better in younger than older participants ( Fig 1B , F(1, 39) = 5.572, p = 0.023, partial χ 2 = 0.125, BF 10 = 2.275), at lower compared to higher set sizes (F(2, 78) = 160.548, p < 0.001, partial χ 2 = 0.805, BF 10 > 100). There was a significant interaction between set size and age group (F(2, 78) = 8.458, p < 0.001, partial χ 2 = 0.178, BF 10 = 59.334), which was driven by larger age-related working memory deficits at set size 2 and 4 (set size 1, F(1, 39) = 0.47, p = 0.497; set size 2, F(1, 39) = 6.18, p = 0.017; set size 4, F(1, 39) = 10.39, p = 0.003). Reaction time (RT) increased with set size ( Fig 1B ; F(2, 78) = 52.359, p < 0.001, partial χ 2 = 0.573, BF 10 > 100) and older participants were slower than younger participants (F(1, 39) = 8.788, p = 0.005, partial χ 2 = 0.184, BF 10 = 7.011). There was no age × set size interaction on RT (F(2, 78) = 1.655, p = 0.198, BF 10 = 0.299).

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

( A ) Delayed match-to-sample task. One, 2, or 4 images were presented sequentially. Intertrial interval was jittered from a uniform distribution (1.2 to 1.6 s). ( B ) Behavioral results. There is a pronounced age-related memory accuracy decrease at higher loads. No significant age × set size interaction effect was observed in reaction time. Lighter and darker colors represent younger and older adults, respectively. Error bars show standard error of the mean. Circled dots show individual participant data. * p < 0.05, ** p < 0.01. Source data can be found at https://doi.org/10.5281/zenodo.12735828 ( S1 Data ).

https://doi.org/10.1371/journal.pbio.3002784.g001

Beta variability tracks brain dynamics as hypothesized by cognitive neurophysiological theory of aging

Changes of rhythmic activity at the trial level lead to alterations in trial-averaged power and cross-trial variability ( Materials and methods , S1A Fig ). Rhythmic activity, unless stated otherwise, was measured using cross-trial variability, which captures fluctuations unique to each individual. Cross-trial beta band variability captured the load-dependent neurophysiological changes in younger and older adults predicted by the Compensation-Related Utilization of Neural Circuits Hypothesis (CRUNCH) [ 22 ]. CRUNCH posits that older adults would not show a parametric neural change with memory load increases during maintenance. This is because older adults overrecruit resources at low set sizes resulting in a resource shortage when set size further increases. We selected the frontal and centroparietal clusters as the channels of interest due to their relevance to working memory function as revealed in previous studies [ 23 , 24 ].

When examining beta variability of each set size and age group during the maintenance phase, we observed an interaction between age group and set size in the frontal cluster alone ( Fig 2A , F(2, 117) = 3.886, p = 0.023, partial χ 2 = 0.087, BF 10 = 40.957; see S1 Table for centroparietal cluster). This suggests that age differentially influences how beta variability changes with memory load. To further quantify this critical interaction effect, we performed linear regression on the beta variability across the 3 set sizes for each participant and tested the slope of their best fit lines at the population level. The parametric variability change with increasing memory load was significant in younger (mean slope = −0.073, t(19) = 3.642, p = 0.002, Cohen’s d = 0.814, two-tailed), but not older participants (mean slope = −0.024, t(20) = 1.474, p = 0.156). In other words, while there was a load-dependent variability decrease in younger adults, older adults failed to show such a systematic modulation. A closer examination of the beta variability modulation pattern in older adults suggests an inability to further modulate beta variability when the memory load increased from 2 to 4 (t(20) = 0.674, p bonferroni > 1.000), echoing predictions from CRUNCH. Analyses of the encoding phase did not reveal any significant differences between younger and older adults, ruling out the possibility that the observed group differences in load-dependent changes during maintenance stemmed from the encoding phase ( S1 Fig ). Together, these analyses suggest that beta-band dynamics during maintenance capture the fundamental premises of CRUNCH, and cross-trial beta variability is equipped with the sensitivity for investigating mechanistic differences in memory processes between younger and older adults.

( A ) Averaged beta-band variability at frontal sites during maintenance (0 to 3 s) and post-response (0.1 to 0.5 s). Inserted panel shows mean slope of load-dependent beta variability during maintenance and main effect of set size during post-response. Gray lines show individual data. ( B ) Maintenance beta variability predicts younger adults’ working memory accuracy. The behavioral relevance of maintenance beta variability was weak at load 1 (Younger: Rho pearson = 0.202, p = 0.392; Older: Rho pearson = 0.287, p = 0.202), suggesting that maintenance-related activity was not behaviorally predictive when the task was less demanding. ( C ) Post-response beta variability at the frontal and centroparietal clusters predicts older adults’ memory accuracy. For frontal beta variability, there was an outlier in the older group that showed high increase of beta variability. Excluding the outlier did not change the statistical significance of the correlation between post-response variability and memory accuracy (B = 0.055, p = 0.004, R 2 = 0.072). Shaded regions represent 95% confidence intervals. R 2 represents variance explained by maintenance or post-response beta variability. Source data can be found at https://doi.org/10.5281/zenodo.12735828 ( S2 Data ).

https://doi.org/10.1371/journal.pbio.3002784.g002

Beta variability during the maintenance phase predicts working memory performance exclusively for younger adults

Since variability-based measures are deemed superior for detecting interindividual differences [ 18 ], we leveraged cross-trial variability to assess more fine-grained differences between age groups. To examine whether beta variability during maintenance predicts individual memory performance and whether such an association presents differently between the age groups, we performed a generalized linear mixed model (GLMM), comparing the effect of frontal beta variability estimated across trials of each set size on memory accuracy between younger and older adults (see Materials and methods ). We observed a significant interaction of age group and beta variability on memory performance (F(1, 119) = 4.330, p = 0.040, partial χ 2 = 0.035, BF 10 = 8.545). This suggests a differential relationship between maintenance beta variability and memory performance in younger and older adults. Further analysis revealed that younger participants with lower variability during maintenance performed better ( Fig 2B ; B = −0.075, p = 0.025, R 2 = 0.060), especially when examining set size 2 and 4 (B = −0.137, p = 0.001, R 2 = 0.218). In contrast, beta variability during maintenance failed to predict memory performance for older adults (B = 0.003, p = 0.905). This implies that beta variability during the maintenance phase not only showed load-dependent changes at the population level but also predicted interindividual differences in memory performance selectively for younger adults. Aging, on the other hand, appeared to impede these systematic modulations to an extent that behavioral relevance of interindividual beta variability was no longer evident in older adults. The absence of maintenance-related activity in predicting older adults’ performance suggests that, while maintenance is influenced by aging (as evident in CRUNCH-like observations reported above), it may not be the primary working memory processing component that predicts behavioral differences in older adults at the individual level. In light of this finding, we investigated age-related differences beyond the maintenance phase.

Beta variability during the post-response phase predicts working memory performance exclusively for older adults

Given that maintenance-related beta variability could not track interindividual differences in older adults, we hypothesized that the post-response phase may capture such differences. This hypothesis was derived from 2 premises. One, the inhibition deficit theory implicates deletion deficits to be the primary driver of age-related memory decline [ 8 ]. Two, beta rhythmic dynamics post-response, particularly originating from where memory representations are maintained, have been interpreted as the neurophysiological signal of memory deletion [ 13 , 25 ]. Together, these premises link post-response beta rhythms with working memory deficits in aging. To test this hypothesis, we first examined whether frontal beta variability showed systematic changes with set size post-response (0.1 to 0.5 s). Unlike the maintenance phase where such a systematic change was evident only in younger adults, we found that beta variability significantly reduced with increasing set sizes for both younger and older adults ( Fig 2A ; F(2, 117) = 4.852, p = 0.009, partial χ 2 = 0.133, BF 10 = 98.128), with no significant difference between groups (F(1, 117) = 0.319, p = 0.573, BF 10 = 1.172) or age × set size interaction effect (F(2, 117) = 0.658, p = 0.520, BF 10 = 1.921). These findings suggest that post-response beta dynamics continue to be associated with working memory function despite aging.

Next, we examined whether post-response beta variability predicted individual memory performance of each set size, using a similar GLMM as was performed for the maintenance phase. We found a surprising reversal of patterns relative to those observed during the maintenance phase. Post-response frontal beta variability correlated with memory performance in older but not younger adults ( Fig 2C ; Younger: B = 0.005, p = 0.847; Older: B = 0.040, p < 0.001, R 2 = 0.094; age × beta variability interaction, F(1, 116) = 4.155, p = 0.044, partial χ 2 = 0.034, BF 10 = 5.972). In other words, while individuals’ memory performance for younger adults was driven primarily by frontal beta variability during maintenance, it was the post-response variability that determined memory performance for older adults. A similar pattern was observed for centroparietal channels (age group × beta variability, F(1,119) = 16.736, p < 0.001, partial χ 2 = 0.127, BF 10 > 100; Older: B = 0.057, p < 0.001, R 2 = 0.186; Younger: B = −0.035, p = 0.086). Given the temporal progression between maintenance and post-response phases, one may consider the individual correlation results in older adults as a later manifestation of a maintenance-related effect, perhaps due to overall slowing of information processing with aging [ 26 ]. However, this possibility is ruled out when examining the direction of the association between beta variability and memory performance. While reduced variability during maintenance predicted better performance for younger adults, it was increased variability during the post-response phase that predicted better performance for older adults. The dynamic change of beta variability during maintenance and post-response matches observations from previous research [ 13 ]. Our findings suggest that older and younger adults differentially leverage beta dynamics during distinct working memory processes to optimize their memory performance. Regulating beta variability during maintenance benefits younger adults but with aging, and perhaps due to the structural and functional reorganization that accompanies it [ 27 ], the neurophysiological locus predicting interindividual working memory differences in older adults shifts to the post-response phase.

With the post-response phase emerging as the possible locus of memory predictive activity in older adults and previous evidence for marked functional reorganization with aging [ 27 ], we examined brain networks recruited during the post-response phase, which may underlie the different contributions of beta variability between younger and older adults. Younger adults, whose memory performance was not associated with post-response beta variability at the individual level, recruited a widely distributed network spanning frontal and centroparietal brain regions ( Figs 3A and S2 ) [ 10 ]. On the other hand, older adults showed, on average, a spatially restricted network confined to centroparietal regions with a marked absence of frontal engagement. We speculated that the sensitivity of beta variability to individual memory performance in older adults may be related to the extent to which an older individual is able to recruit the frontal cortex during the post-response phase. Indeed, the individual with the best memory performance exhibited pronounced increases in beta variability in frontal regions relative to the participant with median memory accuracy. In contrast, the individual with the lowest memory accuracy did not show any increase in beta variability in any region. While we consider these results qualitative and preliminary, they suggest that the deficient ability to involve the frontal beta rhythms may lead to memory decline with aging. This also contributes to recognizing post-response beta variability as the sensitive index that tracks interindividual differences in older adults.

( A ) Source reconstruction of post-response beta variability. Top panel shows cluster-based permutation t -values ( p < 0.005) when comparing post-response beta increase relative to pre-response (−0.4 to −0.1 s). Exemplars represent older participants in the 99%, 50%, and 1% percentile of the present sample based on averaged memory accuracy. ( B ) Single-trial analysis. Stronger frontal beta activity benefits memory accuracy of the next trial. Violin plot shows the distribution of trial-wise post-response beta power of trial n-1. Red crosses represent the mean. Line plot shows the averaged data. Error bars represent standard error of the mean. ( C ) Correct trials showed larger post-response beta variability increase. Colored dots represent individual participant data. Source data can be found at https://doi.org/10.5281/zenodo.12735828 ( S3 Data ).

https://doi.org/10.1371/journal.pbio.3002784.g003

Post-response beta dynamics likely index information deletion: Evidence from 2 converging analyses

So far, we have observed a dissociation in the working memory phase where cross-trial variability in the beta band predicts memory performance in younger and older adults.

As previously mentioned, changes in post-response beta activity, indexed by beta bursting or beta power, have been understood as a signature of information deletion [ 13 , 28 ]. Since cross-trial variability is computed using power data from single trials, it is thus possible that cross-trial beta variability also characterizes the deletion process. Indeed, we found power changes at the single-trial level were associated with changes in cross-trial variability ( S1A Fig ), suggesting power and variability changes likely arise from the same cognitive process in this context. To explore this possibility, it is important to verify whether beta activity measured in any form provides collective evidence of information deletion. To this end, we examined how single-trial power, upon which cross-trial variability is computed, impacted memory performance. Then, we extended this analysis and directly examined the association between cross-trial variability and memory performance.

First, we speculated that post-response beta power in each trial should influence the accuracy of the next trial. Specifically, if post-response beta increase indexes deletion of memory representations, then stronger beta power at the single-trial level should benefit performance in the next trial. Consequently, there should be a positive correlation between post-response beta power of the current trial and memory performance of the next trial (brain-to-behavior relation, an “N+1 correlation”). Indeed, larger post-response beta power increase in the previous trial facilitated performance of the current trial exclusively for older adults ( Fig 3B ; Older: t(8832) = 5.662, p < 0.001, Cohen’s d = 0.175; Younger: t(9053) = 0.586, p = 0.557, permutation t test, two-tailed, alpha = 0.001; age groups × post-response beta power of previous trial, F(1, 17817) = 4.296, p = 0.038, log odds ratio = 0.458), see S2 Table and S3 Fig ). This pattern of results fits the deletion account, suggesting that a stronger increase in beta power post-response potentially frees up the capacity-limited working memory, thereby benefitting the next trial’s performance.

The preceding analysis rests on power changes in single trials. Does cross-trial variability computed using single-trial power also show evidence in favor of the deletion account? If beta variability also indexes deletion, then the extent to which variability is modulated in the post-response phase might be determined by the strength of the memory representations. When memory representations are weak or partial such that they produce an erroneous response, less deletion will be required on those incorrect trials relative to correct trials. Thus, post-response beta variability should be smaller on incorrect trials relative to correct trials irrespective of age group (behavior-to-brain relation). This was indeed the case. Incorrect trials showed lower frontal beta variability than correct trials ( Fig 3C ; F(1, 38) = 7.052, p = 0.012, partial χ 2 = 0.157, BF 10 = 3.923). These effects were consistent across both younger and older participants, with no significant difference between age groups (F(1, 38) = 0.615, p = 0.438, BF 10 = 0.451), and no interaction effect of age and correctness (F(1, 38) = 2.106, p = 0.155, BF 10 = 0.794). Thus, post-response frontal beta dynamics were more pronounced in correct trials where the strength of memory representations was stronger and require more deletion.

Both analyses provide converging findings. The behavioral association of single-trial beta power and cross-trial variability fits the predictions of the deletion account of beta rhythms. This provides confidence in our understanding that the specific prediction of interindividual differences in memory performance of older adults during the post-response phase by variability in the beta band likely reflects a facet of the deletion process.

Control analyses: Ruling out alternative explanations for the post-response beta increase

We have excluded several competing explanations of the post-response beta effect. First, the increase in post-response beta variability should not be interpreted as a reflection of error monitoring. Stronger monitoring is typically reported in incorrect trials compared to correct trials [ 29 ]. However, we observed the opposite trend in post-response beta variability, with a more substantial increase found in correct rather than incorrect trials. Moreover, error monitoring signals have typically involved lower frequency activity in the theta band [ 30 ], unlike the beta-band frequency under consideration here. As such, beta variability increase is a poor candidate for post-response error monitoring.

Second, increase in post-response beta activity could be interpreted as a feedforward confidence estimation process [ 31 ], which plausibly explains the stronger beta variability increase on correct trials relative to incorrect trials, overall. To address this possibility, we again leveraged single-trial beta power upon which variability measures are based. This time, however, we examined the behavioral correlation of single-trial beta power on current trial performance as done in previous studies [ 31 ]. Specifically, if beta power indexes confidence, then it should positively correlate with memory performance in the same trial (an N-N correlation). However, this was not the case. Behavioral performance of the current trial could not be explained by post-response beta activity at the trial level (accuracy: F(1, 17877) = 1.276, p = 0.259; RT: F(1, 15602) = 0.894, p = 0.345), and current trial’s post-response beta power was not modulated by accuracy (F(1, 17877) = 1.044, p = 0.307) or RT (F(1, 15602) = 1.844, p = 0.174). Thus, we do not consider confidence-related processing, characterized by post-response beta power increase, to be a compelling explanation in this case. Consequently, we do not consider changes in variability, stemming from power changes, to be reflective of the confidence estimation process either.

Lastly, there is a possibility that the observed N+1 correlation reflects the implementation of a preparatory attentional state rather than deletion of previously held memory representations. However, attentional deployment is typically associated with alpha-band activity [ 32 , 33 ]. Control analyses on post-response alpha activity did not show any significant effects (ps > 0.168). In addition, previous literature suggests that desynchronization of alpha rhythms benefits next trial performance [ 32 ]. The direction of this association is opposite in our data where it is enhanced power and variability in the beta band that are reflective of better performance in the following trials. Moreover, given that feedback was presented after responses, preparatory attention in service of the next trial is unlikely to be implemented until feedback offset and the onset of the intertrial interval.

Taken together, we believe that the deletion account of the increased post-response beta power and variability reflects a more coherent and parsimonious explanation than other accompanying cognitive processes during the same information processing window.

Control analyses: Spectral specificity of the dissociable effects

The dissociable effects we observed as to which memory processing phase contributes to individual performance differences in younger and older adults were specific to the beta band. The critical interaction between age and set size during the maintenance phase was not significant at any other frequency outside the beta band ( S1 Table ; ps > 0.126). We further examined other frequencies in the post-response phase. Again, we did not observe any significant differences between younger and older adults outside the beta band (GLMM age × variability, ps > 0.108). Moreover, the differences we observed during both maintenance and post-response phases could not be explained by signal quality, as there was no significant between-group difference in signal-to-noise ratio (SNR) in any frequency band (ps > 0.101). Thus, differences in how cross-trial variability predicts performance during separate working memory stages in younger versus older adults are spectrally specific to the beta band and cannot be explained by nonspecific changes in the EEG signal.

Control analyses: Trial-averaged power modulations

Since cross-trial variability is computed using measures of power, and since cross-trial variability results align with those observed using single-trial power data (for instance, N+1 correlations above), we asked whether examining variability provides any additional benefits over and above the examination of conventional trial-averaged power measures alone. It turned out that trial-averaged power failed to capture several significant observations evident through the examination of variability. First, trial-averaged beta power did not show the critical age × set size interaction during maintenance (F(2, 117) = 2.009, p = 0.137). Moreover, trial-averaged frontal beta power did not reflect the dissociable interindividual correlation between maintenance-related beta activity and memory performance (Younger: B = −0.022, p = 0.316; Older: B = −0.002, p = 0.912). Analyses of the trial-averaged post-response spectral activity failed to establish the frequency-specific behavioral relevance of post-response beta activity. Despite the significant correlation between older adults’ memory accuracy and trial-averaged frontal beta power (Older: B = 0.022, p < 0.001, R 2 = 0.117), this nonspecific correlation was also observed in the delta band (Older: B = −0.015, p = 0.038, R 2 = 0.038), alpha band (Older: B = 0.018, p = 0.007, R 2 = 0.062), and gamma band (Older: B = 0.030, p = 0.044, R 2 = 0.036). Post-response beta-band SNR showed no significant difference between age groups (F(1, 39) = 0.042, p = 0.838), neither did the trial-averaged beta power (F(1, 39) = 0.052, p = 0.820). Of note, the absence of age differences in SNR or trial-averaged power during post-response phase suggests that our results should not be explained by general differences between younger and older adults in the robustness of evoked neural responses to events, which could potentially mark a relevant boundary in a trial. The presence of a significant N+1 correlation using single-trial power together with the absence of age-related maintenance effects and the absence of spectral specificity of post-response beta activity using trial-averaged power imply that trial-wise fluctuations are canceled out in the trial-averaging approach. Thus, cross-trial variability, rather than trial-averaged power, in the beta band, appears to be a trait-like signature that is more sensitive to age-related differences in distinct working memory functions. This agrees with prevailing ideas about the greater sensitivity of variability-based measures in capturing interindividual differences [ 17 – 19 ].

Working memory function is a critical cognitive ability that deteriorates with age following adulthood [ 34 ]. But whether different processes within working memory are differentially affected by age remains understudied. This effort is further complicated by the fact that the degree of deterioration is variable across people [ 2 ]. Explaining the neurophysiology of working memory decline in aging requires us to examine the constituent processes within working memory simultaneously and consider variability at the interindividual and intraindividual levels as a parameterized function to be explained, rather than mere noise [ 16 ]. To this end, we adopted a novel approach to assess between- and within-group differences across ages. We combined cross-trial variability, which has largely been studied with broad-band EEG signal and fMRI hemodynamic responses [ 16 , 17 , 19 ], with rhythmic dynamics in the beta range, and examined them during both working memory maintenance and post-response deletion phases. Our novel analytical approach suggests that, when considering cross-trial fluctuations of beta power, variability explains individual differences in working memory performance during distinct phases for each age group. Whereas individual memory performance of younger adults was explained by frontal beta variability during maintenance, memory performance of older adults was primarily explained by post-response beta variability. Thus, task-related cross-trial variability augments individual state-dependent characteristics and predicts behavioral differences within and across age groups. With the age-related dissociations between maintenance and post-response phases, beta variability may serve as an age-related, task-sensitive signature of individual differences in distinctive working memory computations.

When developing models of age-related decline in working memory, it is imperative to incorporate the cognitive and neural dynamics during each information processing state. Most theories of aging are coarse-grained at the cognitive level of analysis [ 26 ], with little to say regarding distinct information processing phases. As an example, CRUNCH offers a plausible explanation for the pattern of neural effects during maintenance but does not directly address differences in post-response deletion. Our findings provide some relevant insights. For instance, the pattern of results during the post-response stage in the present study suggests impairments in information deletion and a putative inability to recruit compensatory resources. Specifically, CRUNCH would predict the involvement of additional neural processes for rescuing impaired information deletion. This should result in the characteristically saturated neural response pattern with increasing set sizes during the post-response phase, as we observed during the maintenance phase. However, we found consistent positive correlations between post-response beta variability and memory performance across all set sizes. CRUNCH would also predict an overactivation of frontal beta activity or additional engagement of task irrelevant regions during post-response phase to achieve efficient deletion. In contrast, our preliminary source estimation analyses suggest an underrecruitment of frontal regions during the post-response phase. These findings align with the inhibition deficit theory but suggest that compensatory resources, as posited by CRUNCH, could not be instantiated by older adults at least in the present investigation. Perhaps the inability to remove information efficiently during the deletion phase creates a bottleneck. This bottleneck could then influence memory maintenance in the following trial, where compensatory mechanisms during maintenance can still be called upon. In this manner, inefficient information deletion may be one of the reasons for the engagement of compensatory mechanisms during the maintenance phase. By viewing working memory as an information processing system that needs to be continuously regulated, we may be able to bridge the inhibition deficit theory with CRUNCH, through examination of the interdependent nature of information removal and maintenance, as demonstrated in the present study.

We interpret the change in post-response beta dynamics as a reflection of a memory deletion process in agreement with previous studies [ 13 , 28 ]. This interpretation is further supported by the observation of single-trial post-response beta power influencing the memory performance in the next trial. It is possible that changes in single-trial beta power are indices of memory deletion, with cross-trial variability, computed using single-trial power measures, reflecting a trait-like ability to execute and adjust the deletion process when memory needs to be regulated rapidly over trials with varying memory loads. We further think that the overall pattern of results sets the stage for elucidating the nature of the deletion process with greater functional specificity. For instance, it is possible that the increases in post-response beta power and variability signify the demand for deletion (the demand account). This account hypothesizes that stronger beta engagement reflects the absolute amount of information to be deleted. In other words, a stronger increase in trial-wise post-response beta power would reflect a stronger demand for deletion. And since cross-trial variability is computed from, and associated with single-trial power ( S1A Fig ), this relationship may be evident with cross-trial variability also. As a result, this account would expect a larger increase in beta variability with increasing set size. For instance, it is possible that the increase in post-response beta power signifies the demand for deletion. This account hypothesizes that stronger beta engagement (both in terms of single-trial power and cross-trial variability) reflects the absolute amount of information to be deleted. Therefore, this account would expect a larger increase in beta variability with increasing set size. This was not the case in our data where we observed a decrease in post-response beta variability with increasing set size ( Fig 2A , right). The overall pattern of results can be better explained if we consider beta engagement as a reflection of the efficiency of the deletion process (the efficiency account). The efficiency of deletion may be a composite of the total amount of information to be removed, the state and strength of to-be-deleted memory representations, the time available for the deletion process, and the rate at which information can be deleted. Given a fixed period of time available for uninterrupted deletion (500 ms in the present work), a smaller proportion of information could be removed when the total amount of information to be removed is higher (for instance, load 4) than lower (for instance, load 1). The negative association between beta variability and set size might suggest that a smaller proportion of to-be-removed information has been removed in the window of analysis at higher set sizes compared with lower set sizes. To the extent that information can be removed more efficiently within the same time window, deletion will be facilitated and performance on the next trial will benefit. This explains why we observe the N+1 correlations whereby a stronger beta power in the present trial benefits performance on the following trial, even though, overall, the efficiency of the deletion process may be lower at set size 4 given a limited deletion period. New experiments are needed to test these novel interpretations of beta dynamics. Mapping out the deletion dynamics over post-response periods of different lengths [ 35 ] and various memory loads [ 10 ] may be a good starting point.

The present findings set the stage for multiple future investigations. For example, it would be prudent to examine the role of object familiarity in the integrity of maintenance and deletion processes. Older people may exhibit some differences in their ability to recall and name objects [ 36 – 38 ], which could at least partially contribute to some memory differences in the present findings. Replicating the study while measuring levels of familiarity and comparing performance with conditions involving abstract, nonnameable objects may be one such approach. It would also help to replicate the findings with larger sample sizes. While the present study was adequately powered to detect an interaction effect between age and set size on memory performance, the interindividual correlations emerging from the present investigation can now be subjected to further follow-up investigations with a larger sample size. In addition, given recent reports suggesting changes in instantaneous beta frequency on a trial-by-trial basis [ 39 ], granular investigations on the relationship between deletion and trial-wise or individualized peak frequencies can be implemented. Our findings also hold implications for cognitive processes beyond those being investigated in the present study. For instance, it would be interesting to see whether the beta rhythmic dynamics facilitating deletion in the present study also contribute to other regulatory processes such as deprioritization [ 40 ], directed forgetting [ 41 ], or controlled removal operations such as information suppression or replacement [ 6 ]. Moreover, whether similar neural mechanisms contribute to the transfer of information between working memory and long-term memory needs to be investigated, for it may hold the key to understanding how representations of our continuous experience in working memory are transformed into discrete, segmented representations in long-term memory [ 42 , 43 ]. Finally, future research is needed to test the causal role of beta activity in modulating the influence from deficient deletion to subsequent maintenance in older adults. It may turn out, as our results showing age-related dissociations between maintenance and post-response indicate, that working memory in younger and older adults may have distinct influences of different neural mechanisms in influencing memory performance, suggesting new directions for future model building, and, ultimately, a more comprehensive mechanistic understanding of cognitive aging in health and disease.

Materials and methods

Ethics statement.

The study protocol was reviewed and approved by the Boston University Institutional Review Board (IRB number 4230E). The research adhered to the ethical guidelines outlined in the Declaration of Helsinki. Written informed consent was obtained from all participants. Participants were compensated $15 per hour.

Participants

Power analysis (80% power, p = 0.05; repeated measures for the critical age × set size interaction) on pilot data ( n = 10) showed a Cohen’s f effect size of 0.248 for the interaction effect, indicating that a total sample size of 28 participants (14 participants per group) would be sufficient to reliably capture an effect. To account for dropouts and exceed these minimum power calculations, we sought at least 20 participants per age group. Twenty-two younger adults and 21 older adults from the greater Boston metropolitan area enrolled in the study. All older participants were prescreened either via phone or an online questionnaire to ensure study eligibility on the following criteria: (1) normal or corrected-to-normal vision and hearing; (2) fluent English speaker; (3) no history of neurological problems or head injury; (4) never been knocked unconscious for longer than 10 minutes; (5) not currently pregnant during the time of study participation; (6) no metal implanted in the head; (7) no implanted electronic devices (pacemaker, neurostimulator); (8) no formal diagnosis of severe tinnitus; (9) no formal diagnosis of a substance problem (related to alcohol or drugs of any kind). Two younger participants’ data were excluded from the analysis due to excessive eye blinks (>60% trials removed in preprocessing). The final sample consisted of 20 younger participants aged 19 to 27 years old (22.1 ± 2.5 years, 10 females, education years 16.0 ± 2.3) and 21 older participants aged 60 to 81 years old (70.6 ± 4.8 years, 7 females, education years 17.3 ± 2.9). Three older participants performed 20 blocks, 23 blocks, and 26 blocks out of a total of 30 blocks due to fatigue.

Behavioral task

We used object stimuli from a previous study [ 44 ]. One, 2, or 4 images of objects were presented sequentially in each trial. Each object was presented for 200 ms followed by an interstimulus interval (ISI) of 1,000 ms. Once all stimuli were presented, the fixation cross turned green. After a delay of 3,000 ms, a probe image was presented for 200 ms, and participants were asked to determine whether the probed image was either identical (50%) or different (50%) from the previous images, by pressing one of 2 buttons on a handheld gamepad. Participants had unlimited time to respond but were instructed to respond as quickly and accurately as possible. Feedback was presented after 500 ms of the response, in the form of a colored circle for 1,000 ms. Yellow indicated a correct response and blue indicated an incorrect response. Color mapping was counterbalanced across participants. New trials began after a jittered time of 1,200 to 1,600 ms (uniform distribution). There were 30 blocks, each containing 24 trials with mixed set sizes, resulting in a total of 720 trials evenly divided among the 3 different set sizes. Participants completed multiple practice blocks until they understood the instruction, showed an averaged accuracy above 0.8, and felt comfortable proceeding.

Electroencephalography

EEG was recorded at a sampling rate of 1,000 Hz in a dimly lit EEG booth using 64 Ag/AgCl electrodes mounted in a BrainCap elastic cap according to the international 10–20 system. The right mastoid electrode served as the online reference. Data were online bandpass filtered to 0.01 to 125 Hz. Four EOG channels were placed at the outer canthi of each eye, above and below the left eye. Impedance levels were kept below 10 kOhm. Participants were instructed to fixate on the central cross throughout each trial, minimize eye blinks and facial movements, and remain still during each block.

EEG preprocessing

EEG preprocessing and analysis were conducted using custom Matlab scripts with the EEGLAB [ 45 ] and Fieldtrip [ 46 ] toolboxes. Raw EEG data were bandpass filtered from 0.5 to 40 Hz and re-referenced to the average of both mastoids. Broken channels were interpolated using a spherical spline method (EEGLAB function, “pop_interp(‘spherical’)”). We extracted epochs time-locked to delay onset (−5.7 to 6 s) and performed independent component analysis (ICA) to correct artifacts caused by eye movements, blinks, heart, muscle, and line noise. The number of removed components was slightly more in older than younger participants (Older: 7.6 ± 2.9; Younger: 6.2 ± 1.6; t(39) = 1.926, p = 0.061). We also removed trials with noisy data points that exceeded a voltage threshold of 100 μV. Improbable and abnormally distributed data points beyond 8 standard deviations of the mean probability distribution and kurtosis distribution were also removed. After preprocessing steps, there were 79.4 ±10.2% clean trials from younger adults and 80.0 ± 9.3% from older adults. Additional segmentation was performed depending on time periods of interest with 0 to 3 s for maintenance and 0.1 to 0.5 s post-response phases.

Behavioral analyses

Mean RT was calculated based on correct trials. Trials with RTs slower than 5 s, or beyond 3 standard deviations of individuals’ own mean RT were excluded from averaging. Two-way mixed Omnibus ANOVA was conducted with the between-participants factor of group (younger versus older) and the within-participants factor of set size (1 versus 2 versus 4). Bonferroni correction was applied for multiple comparisons. Bayesian ANOVA was performed using JASP 0.17.3.

Time-frequency decomposition

For each trial, we subtracted the averaged waveform of each set size to remove phase-locked activity. Single-trial EEG spectral decomposition was then performed for frequencies ranging from 1 to 40 Hz (1 Hz steps) using Morlet wavelets (width linearly increased from 2 to 10) with a time window of 50 ms. Baseline normalization was performed using decibel conversion relative to the pre-trial (−0.4 to −0.1 s) or pre-response period (−0.4 to −0.1 s) for maintenance and response-locked analysis, respectively. SNR was calculated at the averaged power (signal) divided by the standard deviation of the mean across trials (noise) during the interval 0 to 4 s relative to maintenance onset.

Variability analysis

To examine the load-dependent beta variability changes ( Fig 2A ), we performed linear regression on the beta variability across the 3 set sizes for each participant and tested the slope of their best fit lines versus zero at the population level. To assess the interindividual behavioral relevance of neural variability decrease, while controlling for the intraindividual load-dependent effect during maintenance and post-response ( Fig 2B and 2C ), GLMM was formulated as “averaged memory accuracy ~ age group * beta variability + (1 | set size)”. The intercept of each set size was added as the random effect because interindividual differences rather than intraindividual manipulation (set size effect here) is the primary focus. Moreover, we anticipated that individuals with lower variability during maintenance would show better memory performance across all set sizes. Averaged memory accuracy for each set size would theoretically range from 0 to 1 and the accuracy distribution in our case showed negative skewness. Thus, response distribution was specified as “Gamma” with inverse link function. We further confirmed the result by normalizing the response variable (Matlab function “betafit” and “betainv”). With the normalization, the link function was specified as “identity” and the response distribution was specified as “normal”. Regardless of the choices of link function and response distribution, results were consistent (maintenance: F(1, 119) = 3.806, p = 0.053, BF 10 = 6.473; post-response: F(1, 116) = 3.681, p = 0.058, BF 10 = 6.102). Regarding the significant interaction term between age group and beta variability, we constructed a regression model that included set size and beta variability as predictors separately for each age group. Similar analyses were performed on post-response frontal beta variability. An outlier in the older group showing a large post-response frontal beta increase ( Fig 2C ) was identified and excluded from the ANOVA analysis.

Comparison between correct and incorrect trials was performed on load-4 trials to obtain a sufficient number of incorrect trials ( Fig 3C ). Given the unbalanced proportion of correct and incorrect trials, we subsampled from correct trials to avoid bias arising from the unequal number of trials and iterated for 200 times. The averaged results of all iterations were used for statistical tests. An outlier was excluded from the Omnibus ANOVA analysis. Including the outlier did not alter the conclusion (F(1, 39) = 6.177, p = 0.017, partial χ 2 = 0.137, BF 10 = 3.016).

Single-trial analysis

N+1 analysis was performed to reveal the behavioral influence of post-response beta activity on the next trial. Since variability is an aggregated index across trials, we used trial-wise power for this analysis. Single-trial beta power during maintenance (0 to 3 s after stimulus onset) and post-response (0.1 to 0.5 s after response, with the first 100 ms removed to avoid motor artifacts and temporal smearing) were added into the GLMM model with the formula “Memory accuracy (n) ~ maintenance beta power (n) * set size (n) * post-response beta power (n-1) * set size (n-1) * age group + (1 + maintenance beta power + post-response beta power (n-1) + set size (n) + set size (n-1) | participant)”. The response distribution was specified as “Binomial” and the link function was specified as “logit”. We hypothesized that memory accuracy of the current trial was determined by set size and maintenance activity of the current trial, set size of the previous trial as well as the extent to which previous information was deleted. Participants’ intercept and slope variations of the fixed effect were added as the random term. Given the significant interaction effect between age group and post-response beta power of trial n-1, we separated trial n based on response accuracy and compared the post-response beta power of trial n-1 between correct and incorrect trials in each age group using a permutation t test ( Fig 3B , permutation times = 10,000, two-tailed, alpha = 0.001). The same analysis was conducted on within-trial variability calculated as the variance of normalized beta power during the maintenance and post-response phases.

N-N analysis was performed to investigate the relationship between post-response beta power and response confidence at the trial level. We examined whether the post-response beta power of the current trial could explain the same trial’s behavioral performance (“accuracy (n) or RT (n) ~ post-response beta power (n) * group * set size (n) + (1+ post-response beta power + set size (n) | participant)”). When using RT as the response index, the link function was specified as “inverse” for the gamma distribution. Only correct trials were included for the analysis of RT. When using accuracy as the response index, the link function was specified as “logit” for the binomial distribution. Additionally, we modeled the influence of response on post-response beta power (“post-response beta power (n) ~ RT (n) or accuracy (n) * group * set size (n) + (1+set size (n) + RT (n) or accuracy (n) | participant)”) with the link function specified as “identity” for the normal distribution.

Source localization

Linearly constrained minimum variance (LCMV) beamforming was used to reconstruct the cortical sources of post-response neural variability changes. Sensor-level data were referenced to the common average. A standard anatomical MRI and a boundary element method (BEM) headmodel from the Fieldtrip toolbox were used to construct a 3D template grid at 1 cm resolution in Montreal Neurological Institute template space. Given the distance between EEG electrodes and the scalp, we moved the brain surface 5 mm inwards from the skull to accommodate BEM stability (“ft_prepare_sourcemodel”). Channel neighborhood was defined using the default EEG template (“easycapM11_neighb.mat”). A common spatial filter was computed based on band-passed (15 to 25 Hz) EEG time series of all set sizes. Virtual channel time courses for all voxels were reconstructed separately for each set size using the common filter. We then performed time-frequency decomposition and calculated the relative variance on this virtual-channel data as we did on the sensor level. The leadfield comprises 4,050 grids. Source space cluster-based permutation tests were conducted using paired t tests on response relative to the pre-response baseline (−0.4 to −0.1 s). Monte Carlo nonparametric randomization was iterated for 10,000 times with the alpha level of the permutation test set to 0.005.

Supporting information

S1 fig. beta variability during encoding..

Related to Fig 2 . ( A ) Example data for load 1. Gray curves represent trial-wise beta power. The purple curve represents the average of all trials, and the green curve illustrates cross-trial variability. Cross-trial beta variability and trial-averaged beta power showed consistent patterns across time. They were correlated during maintenance (Load 1: Pearson’s Rho = 0. 841, p < 0.001; Load 2, Pearson’s Rho = 0.871, p < 0.001; Load 4, Pearson’s Rho = 0.683, p < 0.001) and post-response phases (Load 1: Pearson’s Rho = 0. 939, p < 0.001; Load 2, Pearson’s Rho = 0.951, p < 0.001; Load 4, Pearson’s Rho = 0.947, p < 0.001). Changes in single-trial beta power affect cross-trial variability. When we median-split beta power of each participant and compared the variance between subset of trials with higher and lower beta power, we found that trials with higher beta power had greater variance than those with lower beta power (maintenance: t(40) = 3.754, p < 0.001, Cohen’s d = 0.747; post-response: t(40) = 1.818, p = 0.038, Cohen’s d = 0.275, one-tailed t test). ( B ) Beta-band variability time course. Vertical dashed lines denote stimulus onsets. ( C ) Item-specific variability during encoding. Data were averaged from variability changes during each stimulus presentation (0 to 1.2 s). ( D ) Slope of variability changes. Boxplot shows the distribution of slopes. Slopes were obtained from linear fitting of beta variability changes induced by consecutive stimulus presentations in both load 2 and load 4. The central line represents the median, and black crosses represent outlier data points beyond 1.5 times the interquartile range. No significant age-related differences were observed in the slopes (Load 2: t(39) = 1.054, p = 0.298; Load 4: t(39) = 1.052, p = 0.299). Source data can be found at https://doi.org/10.5281/zenodo.12735828 ( S4 Data ).

https://doi.org/10.1371/journal.pbio.3002784.s001

S2 Fig. Cross-trial neural variability during maintenance and post-response.

Related to Figs 2 and 3 . ( A ) Time-frequency map of neural variability at frontal sites. ( B ) Frontal beta variability during maintenance phase. Time frequency (difference) map shows frontal neural variability averaged across set sizes. Topographical plots show the averaged beta-band variability (15 to 25 Hz) during the maintenance interval (0 to 3 s). Black dots on the topography highlight the frontal channels used to generate time-frequency maps. Shaded error bars on the time series represent the between-participant standard error. The colored vertical solid lines on the x-axis correspond to the mean RT of each set size. ( C ) Time-frequency (difference) map of post-response neural variability changes at the frontal site. ( D ) Topography of post-response (0.1 to 0.5 s) beta variability increases. The highlighted channels passed cluster-based permutation tests (alpha = 0.001, two-tailed).

https://doi.org/10.1371/journal.pbio.3002784.s002

S3 Fig. Related to Fig 3 .

The influence of post-response beta power of previous trials on memory accuracy. ( A ) The interaction effect among set size of the previous trial, post-response beta power of the previous trial, and the current trial’s set size. ( B ) The interaction effect of set size of previous trials, post-response beta power of the previous trial, and age groups. ( C ) The interaction effect among set size of previous trials, post-response beta power of the previous trial, current trial’s set size, and age groups. Error bars represent standard error of the mean. Source data can be found at https://doi.org/10.5281/zenodo.12735828 ( S3 and S5 Data).

https://doi.org/10.1371/journal.pbio.3002784.s003

S1 Table. Related to Fig 2 .

Analysis of maintenance-related activity. The critical interaction between age group and set size was observed in the beta band (15–25 Hz). The frequency band of interest was guided by existing literature; however, we also explored other frequency bands to demonstrate frequency specificity. While other spatiospectral combinations did not show significant age × set size interaction effects, we proceeded to separate age groups and examined interindividual correlation between maintenance activity and memory accuracy for completeness.

https://doi.org/10.1371/journal.pbio.3002784.s004

S2 Table. Related to Fig 3 .

GLMM output table of single-trial N+1 analysis using the formula “Accuracy (n) ~ set size (n) * maintenance beta power (n) * set size (n-1) * post-response beta power (n-1) * age group+ (1 + set size (n) + maintenance beta power (n) + set size (n-1) + post-response beta power (n-1) | participant)”. Participants’ accuracy of the current trial (n) is determined by set size and maintenance beta power of the current trial n, as well as set size and post-response beta power of the previous trial n-1.

https://doi.org/10.1371/journal.pbio.3002784.s005

S1 Data. Source data of Fig 1 .

https://doi.org/10.1371/journal.pbio.3002784.s006

S2 Data. Source data of Fig 2 .

https://doi.org/10.1371/journal.pbio.3002784.s007

S3 Data. Source data of Fig 3 .

https://doi.org/10.1371/journal.pbio.3002784.s008

S4 Data. Source data of S1 Fig .

https://doi.org/10.1371/journal.pbio.3002784.s009

S5 Data. Source data of S3 Fig .

https://doi.org/10.1371/journal.pbio.3002784.s010

Acknowledgments

We thank Phillip (Xin) Cheng for insightful discussion and Rutvi Jain for assistance with participant recruitment.

• View Article
• Google Scholar
• PubMed/NCBI
• 8. Hasher L, Zacks RT, May CP. Inhibitory control, circadian arousal, and age. In: Gopher D, Koriat A, editors. Attention and performance XVII: Cognitive regulation of performance: Interaction of theory and application. The MIT Press; 1999. pp. 653–675.

June 5, 2017

Working Memory: How You Keep Things “In Mind” Over the Short Term

Given its central role in our mental life working memory may become important in our quest to understand consciousness itself

By Alex Burmester & The Conversation US

Ben Pipe Photography Getty Images

The following essay is reprinted with permission from  The Conversation , an online publication covering the latest research.

When you need to remember a phone number, a shopping list or a set of instructions, you rely on what psychologists and neuroscientists refer to as working memory. It’s the ability to hold and manipulate information in mind, over brief intervals. It’s for things that are important to you in the present moment, but not 20 years from now.

On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing . By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.

Researchers believe working memory is central to the functioning of the mind. It correlates with many more general abilities and outcomes—things like  intelligence  and  scholastic attainment —and is linked to basic sensory processes.

Given its central role in our mental life, and the fact that we are conscious of at least some of its contents, working memory may become important in our quest to understand consciousness itself. Psychologists and neuroscientists focus on different aspects as they investigate working memory: Psychologists try to map out the functions of the system, while neuroscientists focus more on its neural underpinnings. Here’s a snapshot of where the research stands currently.

How much working memory do we have?

Capacity is limited—we can keep only a certain amount of information “in mind” at any one time. But researchers debate the nature of this limit.

Many suggest that working memory can store a  limited number of “items” or “chunks” of information . These could be digits, letters, words or other units. Research has shown that the number of bits that can be held in memory can depend on the type of item—flavors of ice cream on offer versus digits of pi.

An alternative theory suggests working memory acts as a  continuous resource  that’s shared across all remembered information. Depending on your goals, different parts of the remembered information can receive different amounts of resource. Neuroscientists have suggested this resource could be  neural activity , with different parts of the remembered information having varying amounts of activity devoted to them, depending on current priorities.

A different theoretical approach instead argues that the capacity limit arises because different  items will interfere with each other in memory .

And of course memories decay over time, though rehearsing the information that’s in working memory seems to mitigate that process. What researchers call maintenance rehearsal involves repeating the information mentally without regard to its meaning—for example, going through a grocery list and remembering the items just as words without regard to the meal they will become.

In contrast, elaborative rehearsal involves giving the information meaning and associating it with other information. For instance, mnemonics facilitate elaborative rehearsal by associating the first letter of each of a list of items with some other information that is already stored in memory. It seems only elaborative rehearsal can help consolidate the information from working memory into a more lasting form—called long-term memory.

In the visual domain,  rehearsal may involve eye movements , with visual information being tied to spatial location. In other words, people may look at the location of the remembered information after it has gone in order to remind them of where it was.

Working memory versus long-term memory

Long-term memory is characterized by a much larger storage capacity. The information it holds is also more durable and stable. Long-term memories can contain information about episodes in a person’s life, semantics or knowledge as well as more implicit types of information such as how to use objects or move the body in certain ways (motor skills).

Researchers have long regarded working memory as a  gateway into long-term storage . Rehearse information in working memory enough and the memory can become more permanent.

Neuroscience makes a clear distinction between the two. It holds that working memory is related to temporary activation of neurons in the brain. In contrast, long-term memory is thought to be related to physical changes to neurons and their connections. This can explain the short-term nature of working memory as well as its greater susceptibility to interruptions or physical shocks.

How does working memory change over a lifetime?

Performance on tests of working memory improves throughout childhood. Its capacity is a major driving force of cognitive development. Performance on assessment tests increase steadily  throughout infancy ,  childhood and the teenage years . Performance then reaches a peak in young adulthood. On the flip side, working memory is one of the cognitive abilities most sensitive to aging, and performance on  these tests declines in old age .

The rise and fall of working memory capacity over a lifespan is thought to be related to the normal development and degradation of the prefrontal cortex in the brain, an area responsible for  higher cognitive functions .

We know that damage to the prefrontal cortex causes working memory deficits (along with many other changes). And recordings of neuronal activity in the prefrontal cortex show that  this area is active during the “delay period”  between when a stimulus is presented to an observer and when he must make a response—that is, the time during which he’s trying to remember the information.

Several mental illnesses, including  schizophrenia and depression , are associated with decreased functioning of prefrontal cortex, which can be  revealed via neuroimaging . For the same reason, these illnesses are also associated with decreased working memory ability. Interestingly, for schizophrenic patients, this deficit appears  more marked in visual rather than verbal  working memory tasks. In childhood, working memory deficits are linked to  difficulties in attention, reading and language .

Working memory and other cognitive functions

The prefrontal cortex is associated with a wide array of other important functions, including  personality, planning and decision-making . Any decrease in the functioning of this area is likely to affect many different aspects of cognition, emotion and behavior.

Critically, many of these prefrontal functions are thought to be intimately linked to, and perhaps dependent on, working memory. For instance, planning and decision-making require us to already have “in mind” the relevant information to formulate a course of action.

A theory of cognitive architecture, called  Global Workspace Theory , relies on working memory. It suggests that information held temporarily “in mind” is part of a “global workspace” in the mind which connects to many other cognitive processes and also determines what we are conscious of in any given moment. Given that this theory suggests working memory determines what we are conscious of, understanding more about it may become an important part of solving the mystery of consciousness.

Improving your working memory

There is some evidence that it’s possible to train your working memory using interactive tasks, such as simple games for children that involve memory ability. It has been suggested that this training can help improve scores on other types of tasks,  such as those involving vocabulary and mathematics . There is also some evidence that training to beef up working memory can  improve performance for children with specific conditions , such as ADHD. However, research reviews often conclude that benefits are  short-lived and specific to the trained task .

Furthermore, the enhancements found in some of these studies could be due to learning how to more efficiently use one’s working memory resources, as opposed to increasing its capacity. The hope for this kind of training is that we can find relatively simple tasks which will both improve performance not just on the task itself but also transfer to a range of other applications.

This article was originally published on  The Conversation . Read the original article .

• Tools and Resources
• Customer Services
• Affective Science
• Biological Foundations of Psychology
• Clinical Psychology: Disorders and Therapies
• Cognitive Psychology/Neuroscience
• Developmental Psychology
• Educational/School Psychology
• Forensic Psychology
• Health Psychology
• History and Systems of Psychology
• Individual Differences
• Methods and Approaches in Psychology
• Neuropsychology
• Organizational and Institutional Psychology
• Personality
• Psychology and Other Disciplines
• Social Psychology
• Sports Psychology
• Share Facebook LinkedIn Twitter

Article contents

Working memory.

• Tom Hartley Tom Hartley University of York
• , and  Graham J. Hitch Graham J. Hitch University of York
• https://doi.org/10.1093/acrefore/9780190236557.013.768
• Published online: 19 October 2022

Working memory is an aspect of human memory that permits the maintenance and manipulation of temporary information in the service of goal-directed behavior. Its apparently inelastic capacity limits impose constraints on a huge range of activities from language learning to planning, problem-solving, and decision-making. A substantial body of empirical research has revealed reliable benchmark effects that extend to a wide range of different tasks and modalities. These effects support the view that working memory comprises distinct components responsible for attention-like control and for short-term storage. However, the nature of these components, their potential subdivision, and their interrelationships with long-term memory and other aspects of cognition, such as perception and action, remain controversial and are still under investigation. Although working memory has so far resisted theoretical consensus and even a clear-cut definition, research findings demonstrate its critical role in both enabling and limiting human cognition and behavior.

• short-term memory
• serial order
• intelligence

Introduction

The term working memory refers to human memory functions that serve to maintain and manipulate temporary information. There is believed to be a limited capacity to support these functions which combine to play a key role in cognitive processes such as thinking and reasoning, problem-solving, and planning. A common illustration is mental calculation which typically involves maintaining some initial numerical information whilst carrying out a series of arithmetical operations on parts and maintaining any interim results. However, the range of activities that depend on working memory is very much wider than that example might suggest. Thus, perception and action can also depend critically on maintaining and manipulating temporary information, as for instance when identifying a familiar constellation in the night sky, or when preparing a meal.

Information about a stimulus remains available for a few seconds after it is perceived (short-term memory) but without active maintenance it rapidly becomes inaccessible ( Peterson & Peterson, 1959 ; Posner & Konick, 1966 ). Conceptually, working memory extends short-term memory by adding the active, attentional processes required to hold information in mind and to manipulate that information in the service of goal-directed behavior.

The short-term storage required for working memory can be distinguished from long-term memory, which is concerned with more permanent information acquired through learning or experience and includes declarative memory (retention of factual information and events) and procedural memory (underpinning skilled behavior; see Cohen & Squire, 1980 ). Notably, and in contrast to short-term memory, these forms of long-term memory are passive in the sense that, once acquired, memory for facts, events, and well-learned skills can persist over very long periods without moment-to-moment awareness. For example, a vocabulary of many thousands of words, including the relationship between their spoken forms and meanings, can be retained effortlessly over a lifetime. Similarly, once acquired through practice, complex and initially challenging behaviors such as swimming or riding a bicycle can become almost automatic and can be carried out with relatively little conscious control.

In early models of the human memory system (e.g., Atkinson & Shiffrin, 1968 ; see Logie, 1996 ) short-term memory was seen as a staging post or gateway to long-term memory, and it was recognized that it could also support more complex operations, such as reasoning, thus acting as a working memory. Subsequent research has attempted to refine the concept of working memory, characterizing its functional role, limits, and substructure, and distinguishing the processes involved in maintenance and manipulation of information from the storage systems with which they interact.

It has proven difficult, however, to disentangle working memory function from other aspects of cognition with which it overlaps. First, as described in more detail in the section “ Substructure and Relationship to Other Aspects of Cognition ,” many current accounts view the mechanisms of working memory as contributing to other perhaps more fundamental functions such as attention, long-term memory, perception, action, and representation. It is also notable that many informal descriptions of working memory emphasize consciousness and awareness as key features. Intuitively, many working memory functions are accessible to consciousness, and concepts such as mental manipulation, rehearsal, and losing track of information through inattention are subjectively encountered as characteristics of the conscious mind. Of course, by definition, people cannot be subjectively aware of any unconscious contributions to working memory (although they can potentially be inferred from behavior). Some theorists have argued that working memory is central to conscious thought (e.g., Baars, 2005 ; Carruthers, 2017 ), while other empirical researchers have sought to demonstrate nonconscious processes operating in what would typically be considered working memory tasks (e.g., Hassin et al., 2009 ; Soto et al., 2011 ). It is not clear whether, how, or to what extent consciousness is essential for working memory functions, or whether indeed the definition of working memory ought to include, or avoid, aspects of conscious experience. This article steers away from the topic, but the current status of the debate is captured in reviews such as Persuh et al. (2018) . Overall, it is difficult to precisely delineate the boundaries of working memory, whether with other cognitive functions or with consciousness and awareness; in philosophical terms it may not constitute a “natural kind” ( Gomez-Lavin, 2021 ).

These challenges make it difficult to establish a clear-cut and uncontroversial definition of working memory itself, its function, and substructure. Yet it is clear that working memory describes a cluster of related abilities that play a critical role in everyday thinking, placing important constraints on what we can and cannot do. Research on the topic has proved fruitful and although there remain many theoretical controversies about how working memory should be defined and analyzed, these mainly relate to the way in which its operations and substrates can be usefully subdivided, and their interrelationships with other cognitive systems such as those responsible for long-term memory and attention (see Logie et al., 2021 for in-depth discussion).

The following sections begin by identifying relatively uncontroversial characteristics of working memory and its temporal and capacity limits before outlining the main theoretical perspectives on the structure of working memory and its relationship to other forms of cognition. This is followed by a summary of the main experimental tasks and key empirical observations which underpin current understanding. Finally, a brief discussion of the importance of working memory beyond the laboratory is provided.

Temporal Limits

It is broadly agreed that its temporary or labile character is a defining characteristic of working memory. In contrast with established declarative and procedural memories that can be retained indefinitely, recently presented novel information is typically lost after a few seconds unless actively maintained. This active maintenance of short-term memory in order to complete a task is one of the core functions of working memory. As discussed further (see “ Limiting Mechanisms ”), it is less clear how such information is lost over time, or whether forgetting is strictly linked to the passage of time (decay) or merely correlated with it (for example, through an accumulation of interfering information). Nonetheless the vulnerability of short-term memory to degradation over time constrains the uses to which it can be put. Active maintenance processes include rehearsal—covertly subvocalizing verbal material, and attentional refreshing—selectively attending to an item that has not yet become inactive (see e.g., Camos et al., 2009 ). These active processes are themselves limited by the modality and quantity of the stored material, so that for instance subvocal rehearsal is disrupted by speaking aloud at the same time (“articulatory suppression”; Murray, 1967 ), and attentional refreshing can only be directed at a limited number of items in a given period of time ( Camos et al., 2018 ). Even though such active maintenance processes extend the temporal limits of short-term memory, when they do so at the cost of limited attentional resources, this reduces the availability of those resources for other goals.

Capacity Limits

It is also agreed that the limited capacity of working memory is a defining characteristic; in subjective terms, only a limited number of items can be “held in mind” at once. For example, in the classic digit span test of short-term memory capacity, participants are asked to briefly store, and then recall in order, arbitrary sequences of digits of gradually increasing length. In this type of task, accurate performance is typically only possible for very short sequences of up to three or four items beyond which errors of ordering become ever more frequent. Memory span is defined as the sequence length at which recall is correct half the time and is found to be between six and seven for digits, and even less for items such as unrelated words ( Crannell & Parrish, 1957 ). Similar capacity constraints are evident in nonverbal tasks requiring the recall of spatial sequences or the locations or visual properties of objects in spatial arrays. For instance, in the Corsi Block task, participants follow an assessor in tapping out a sequence of blocks in a tabletop array or a sequence of highlighted squares on a computer display. In the standard task, nine blocks are used in a fixed configuration and healthy participants can only recall sequences of around six taps even when tested immediately after presentation ( Corsi, 1972 ; Milner, 1971 ). Such tasks are helpful in identifying the fundamental capacity constraints on short-term memory but working memory capacity is also constrained by the active processes that maintain and manipulate information. This is typically assessed using complex span tasks which measure how many items can be held in mind while carrying out an attention-demanding concurrent task, leading to far lower estimates than simple spans ( Daneman & Carpenter, 1980 ). Similarly, participants show greatly reduced performance on a backward digit span task where mental manipulation is required to reverse the original sequence at recall. (Interestingly the Corsi span is the same in both directions; Kessels et al., 2008 ). Notably, forward and backward digit span and Corsi Block tasks are all used in the clinical assessment of neuropsychological patients as well as in research studies, highlighting the importance of working memory capacity in characterizing healthy and impaired cognitive function.

Just as the temporal limits of short-term memory can be extended by active maintenance processes, its capacity limits can be mitigated through strategic processing. Although it is clear that the number of items that can be stored in working memory is limited, there is some flexibility about what constitutes an item. For example, the sequence “1-0-0” might constitute three digits or might be represented as a single item, “hundred.” The possibility of more efficient forms of coding depends on interactions with long-term memory and can be exploited strategically to extend working memory capacity through “chunking” ( Miller, 1956 ). Thus, for an IT professional, the sequence “CPUBIOSPC” is more easily maintained as the familiar acronyms “CPU,” “BIOS,” and “PC” than as an arbitrary sequence of 10 letters.

While the previous example exploits long-term knowledge, even arbitrary grouping can extend the capacity of working memory, for example, in the immediate serial recall of verbal sequences, performance is improved when items are presented in groups. A spoken sequence of digits like “352-168” (i.e., with a pause between the two groups of digits) is recalled more easily than the ungrouped sequence “352168” ( Ryan, 1969 ). Again, this effect can be deployed strategically, and there is evidence that participants spontaneously group verbal material in memory.

More generally, prior learning and experience can not only expand effective storage capacity but can also contribute to efficient active processing operations. For example, children may initially use a counting-on strategy to perform simple sums such as 2 + 3 = 5, but later typically learn arithmetic number facts that automate such operations, in turn permitting more demanding mental arithmetic to be carried out within working memory ( Raghubar et al., 2010 ). In the extreme, expert calculators may collect extraordinarily large “mental libraries” of number facts ( Pesenti et al., 1999 ). Another powerful strategy for extending working memory capacity is seen in expert abacus operators who in mental calculation are able to use visual imagery to internalize algorithms learned from using the physical device ( Stigler, 1984 ).

Limiting Mechanisms

Despite the clear consensus that limited capacity and duration are defining characteristics of working memory, distinguishing it from other forms of memory and learning, there is less agreement about the mechanisms through which information is limited and forgotten.

In one account, the ultimate capacity limits of the system are determined by its access to a limited number of discrete slots, each of which can be used to hold a chunk of information ( Cowan, 2001 ; Luck & Vogel, 1997 ). However, an alternative and increasingly influential view is that working memory has access to a continuous resource which can be flexibly deployed to support a greater number of chunks or items on the one hand, or greater fidelity and precision on the other ( Bays & Husain, 2008 ; see Ma et al., 2014 for discussion).

The loss of information from working memory over time can similarly be attributed to different mechanisms, although here they do not amount to mutually exclusive models of the same phenomenon. One potential mechanism is decay, assumed to be a fundamental property of the substrate of short-term memory, through which information is lost due to the passage of time alone. In this view the attentional/executive component of working memory is typically deployed to extend its capacity by strategically (but effortfully) refreshing or rehearsing the content of short-term memory before it decays irretrievably. A further potential mechanism is interference. In this account, memory traces are prone to be confused with, or gradually corrupt one another. Several current models incorporate a combination of decay and interference ( Baddeley et al., 2021 ; Barrouillet & Camos, 2021 ; Cowan et al., 2021 ; Vandierendonck, 2021 ), while Oberauer (2021) stands out in rejecting time-based forgetting and maintenance processes, proposing in their place loss due to interference, and requiring a process dedicated to the active removal of outdated information from working memory.

Substructure and Relationship to Other Aspects of Cognition

Because it is linked to such a wide range of cognitive capacities, it can be difficult to clearly distinguish mechanisms of working memory from those of its specialized subcomponents or of general-purpose cognitive mechanisms which contribute to nonmemory functions. There is a broad consensus that working memory involves the interaction of an active process (corresponding to “attention” or “executive control”) with a substrate that can represent the content of memory and thus act as a short-term store. Authors disagree, or are sometimes agnostic, as to the extent to which these components can be usefully subdivided and the degree to which they are uniquely involved in working memory or more generally in cognition. Authors also differ in the emphasis they put on different modalities and tasks. These different emphases may sometimes mask a deeper consensus in which models are complementary rather than incompatible ( Miyake & Shah, 1999 ).

Although the term working memory had already been applied to the use of short-term memory in goal-directed behavior ( Atkinson & Shiffrin, 1968 ), it was the influential work of Baddeley and Hitch ( Baddeley, 1986 ; Baddeley & Hitch, 1974 ), that introduced the separation of attentional control processes (governed by a “central executive”) and short-term storage systems (thought of as “buffers,” i.e., distinct and specialized systems). They further identified a distinction between verbal and visual buffers which were subject to different forms of disruption and appeared to use distinct codes. In particular, verbal information could be stored in a speech-based system (termed the “phonological loop”), in which similar sounding items were more likely to be confused and which was disrupted by concurrent articulation. This work led to the development of the multicomponent model, which subsequently incorporated a richer characterization of the visuo-spatial store (the “visuospatial sketchpad,” see e.g., Baddeley & Logie, 1999 ; Logie, 1995) and, later, an additional store—the “episodic buffer” which holds amodal information and interacts with episodic long-term memory ( Baddeley, 2000 ). The possibility of further substructure within these core components is also recognized (e.g., Logie, 1995 on distinguishing visual and spatial subcomponents; see also Logie et al., 2021 on the possibility of multiple substrates within a multicomponent perspective).

An alternative view, the embedded processes model put forward by Cowan (1999) , is that working memory can be seen as the controlled, temporary activation of long-term memory representations, with access to awareness being limited to three to four items or chunks. A key distinction with the multicomponent model hangs on whether working memory relies on a distinct substrate (as implied by the term “buffer”), or whether the substrate is shared with long-term memory. Oberauer (2002) similarly identifies working memory with activated representations in long-term memory. In this account, the activated region forms a concentric structure within which a subset of individual chunks inside a “region of direct access” compete to be selected as the focus of attention.

Other more recent theoretical accounts have also emphasized the role of attentional control in determining the limits of working memory. For example, Engle (2002) regarded capacity constraints as reflecting the limited ability to control domain-general executive attention in situations where there is the potential for interference among conflicting responses. The time-based resource sharing account ( Barrouillet & Camos, 2004 ) highlights the need to balance the active refreshing of short-term with concurrent processing demands. In this view, constraints arise from the necessary trade-off between maintenance and manipulation, both of which rely on common attentional resources.

Many theoretical approaches to working memory do not follow Baddeley and Hitch in identifying modality-specific substrates for the temporary storage of information and assume instead a unitary system in which many different types of feature can be represented (e.g., Cowan et al., 2021 ; Oberauer, 2021 ). In such accounts, modality-specific phenomena are attributed to differences in the extent to which such features overlap within and between modalities. On the other hand, some authors acknowledge the possibility that there may be many alternative substrates, and that even within a modality further subdivisions may be possible. So, for example substrates supporting memory of verbal/linguistic content might further distinguish auditory-verbal, lexical, and semantic levels of representation ( Barnard, 1985 ; Martin, 1993 ).

Neuroscientific investigations have tended to support the consensus idea of a broad separation between executive and attentional control processes on the one hand, and (often modality-specific) stores on the other, but if anything have highlighted even more extensive overlap of the neural substrate of working memory with other cognitive functions including sensory–perceptual and action–motor representation, and greater granularity and fractionation of function within both storage and control systems. This led Postle (2006) to argue that working memory should be seen as an emergent property of the mind and brain rather than a specialized system in its own right:

Working memory functions arise through the coordinated recruitment, via attention, of brain systems that have evolved to accomplish sensory-, representation-, and action-related functions. ( Postle, 2006 ), p. 23

Even in this view it is clear that the mechanisms of working memory (however they overlap with other cognitive functions) involve the interaction of distinct components (at minimum “attention” is distinguished from sensory/representation and action-related function, and these latter functions may also be further subdivided).

Empirical Investigation and Key Findings

A variety of tasks have been developed to investigate working memory in the laboratory. These tasks, of course, always require participants to briefly retain some novel information, often the identity of a set of items which might be visual (for example, colored shapes) or verbal (digits, words, letters). However, they vary quite considerably in the extent to which they require memory for the structure of the set (such as, for verbal stimuli, their order or the spatial layout of an array of items), the degree to which they place an ongoing or concurrent load on memory and attention, and the precision with which sensory and perceptual properties of the individual items must be represented. An excellent overview of these techniques and associated benchmark findings can be found in Oberauer et al. (2018) .

In an item recognition task, participants determine whether a specific item was in a set (a sequentially presented list or simultaneously displayed array) that they previously studied ( McElree & Dosher, 1989 ). In probed recall , they are provided with a cue that uniquely specifies a given item from a previously presented set, which they are then required to recall ( Fuchs, 1969 ). In free recall tasks, typically employed with verbal stimuli, participants are presented with an ordered list, but are allowed to recall the items in any order ( Postman & Phillips, 1965 ), whereas in serial recall ( Jahnke, 1963 ) they are required to retain the original order of presentation.

The preceding tasks place increasing demands on short-term memory for the structure as well as the content of the presented stimuli, but place relatively little requirement for attention or the manipulation of memory content. To address these aspects of working memory, a range of additional tasks have been developed. In complex span tasks the to-be-remembered items are interleaved with a processing task, placing a greater concurrent load on the attentional system ( Daneman & Carpenter, 1980 ). In the N-back task , items are presented rapidly and continuously, with the participant being required to decide whether each new item repeats one encountered exactly n-items earlier in the sequence; to do this they must not only maintain the order of the previous n-items, but also manage the capacity-limited short-term memory resource as every new item arrives. These demands become increasingly taxing as the value of n increases, again giving an indication of the effects of load on performance or, since it is particularly amenable to neuroimaging, brain activity (see Owen et al., 2005 for review). 1 As mentioned, the manipulation requirements of serial recall can be increased by reversing the order in which items are to be recalled. More involved forms of mental manipulation are explicitly tested in memory updating paradigms ( Morris & Jones, 1990 ), within which, after being presented with an array or description, participants are instructed to carry out a sequence of operations before retrieving the result.

To assess its fidelity over brief intervals, tasks that require memory for detailed properties of the items are useful. In change detection tasks (e.g., Luck & Vogel, 1997 ), participants are required to respond to alterations in the stimulus (typically a visually presented array) between presentation and testing. These alterations can be made arbitrarily small, thus testing the precision of the underlying memory representation. Going beyond recognition -like responses to change, in continuous reproduction or delayed estimation tasks , participants are asked to recall continuous features of the stimuli such as the precise color or orientation of a shape within a previously-studied array (e.g., Bays & Husain, 2008 ). These tasks allow researchers to go beyond the question of whether information is merely retained or lost; they can be used to characterize and quantify the quality of the underlying representation, which in turn can shed light on the potential trade-off between capacity and precision in working memory.

The preceding tasks provide a very useful set of tools for investigating working memory in the laboratory. To investigate the structure and operation of the system, experiments typically manipulate characteristics of the items to be stored, and often employ concurrent tasks devised to selectively disrupt putative components or processes. In their standard forms, the individual items are treated as equally valuable or important, but it is also possible to cue specific items, locations, or serial positions in order to encourage participants to prioritize specific content (e.g., Hitch et al., 2020 ; Myers et al., 2017 ). Improved recall for such prioritized items can then reveal the operation of strategic processes. Overall, such manipulations show a range of replicable effects, not just on overall performance and response times, but also on patterns of error. In turn these benchmark effects have provided the impetus for current theories and provide important constraints for emerging computational models of working memory ( Oberauer et al., 2018 ).

Set Size and Retention Interval Effects

The most important effects relate to capacity and temporal limits that have already been discussed, and these apply across all applicable experimental paradigms and modalities. Specifically, in terms of capacity limits, task accuracy is impaired as the number of items (set size) is increased (response times also generally increase with set size), and in terms of temporal limits, accuracy declines monotonically with the duration of a delay between presentation and testing. The latter effect is reliably seen for both verbal and spatial materials when the retention interval is filled with a distracting task. It does not apply to unfilled delays in tasks with verbal materials, and only sometimes occurs with spatial materials. The difference between filled and unfilled delays forms part of the evidence in favor of the core working memory concept of active executive/attentional processes in sustaining otherwise fleeting short-term memories.

Primacy and Recency Effects

Another signature of working memory is that items are retrieved with greater accuracy if they are presented at the beginning (primacy) or end (recency) of a sequence relative to other items. The operation of primacy and recency effects is seen in immediate serial recall and other tasks where the presentation order is well-defined, and for both verbal and visuo-spatial content. This leads to a serial position curve (in which accuracy is plotted for each serial position in a list) with a characteristic bowed shape. The effect suggests that a shared or general serial ordering mechanism privileges access to these serial positions in an ordered list and/or impairs access to other serial positions. It is important to note that primacy and recency effects are also observed in the immediate free recall of lists of words when the capacity of working memory is greatly exceeded and where they may have a very different explanation (see e.g., Baddeley & Hitch, 1993 ).

Errors and Effects of Similarity

Working memory errors frequently involve confusion between items in the memory set. This is evident in a wide range of tasks (including variants of recognition, change-detection, and continuous reproduction tasks), but is perhaps clearest in immediate serial recall, where the most common forms of error involve the misordering of items. These errors most frequently involve local transpositions in which an item moves to a nearby list position, often exchanging with the item in that position. For example, a sequence like “D, F, E, O, P, Q” might be recalled as “D, F, O, E, P, Q.” Items are most likely to transpose to immediately adjacent list positions, with the probability of a transposition decreasing monotonically as the distance within the sequence increases. Note that there are fewer opportunities for local transpositions at the beginning and end of a sequence so the locality constraint on transpositions likely plays at least some role in primacy and recency effects.

In a verbal working memory task, when items from the memory set are confused with one another, they are most likely to be confused with phonologically similar items making performance for lists of similar sounding items poorer than for phonologically distinct items. In serial recall, this effect manifests itself as an increased tendency for phonologically similar items to transpose with one another, so that in the preceding example, items “D,” “E” and “P” (because they rhyme) would be more likely to transpose with one another than items “F,” “O,” and “Q.” Although these similarity effects are largely reported in verbal paradigms, analogous findings are sometimes observed with visual materials (for example, a sequence of similar colored shapes is harder to reconstruct than a sequence of distinctively colored shapes; Jalbert et al., 2008 ).

The analysis of errors and confusion has been critical in understanding the nature of representation in verbal working memory (for example, demonstrating the importance of speech-based rather than semantic codes), in developing the concept of the phonological loop, and in developing computational models which account for these findings in terms of underpinning serial ordering mechanisms.

Individual Differences and Links With Other Facets of Cognition

Speaking to questions about the relationship between working memory and other aspects of cognition, another set of benchmark findings is concerned with correlations between performance on working memory tasks and other measures. In particular, working memory is correlated with measures of attention and fluid intelligence (the capacity to solve novel problems independent of prior learning; see e.g., Engle, 2002 ) suggesting that all three constructs involve common resources. There is consensus that aspects of attention contribute to working memory, but attention is also relevant to tasks that make minimal demands on memory. At the same time, working memory plays an important role in problem solving in the absence of relevant prior learning, but it can also be applied to tasks that do not involve complex problems. This suggests a hierarchical relationship in which limited cognitive resources (i.e., attention) are applied to maintain and manipulate information in memory (attention + short-term memory = working memory) in the context of demanding problems (working memory + problem solving = fluid intelligence).

This somewhat simplistic sketch of the relationship between constructs omits the contribution of long-term memory and learning to working memory. That contribution is evident in several empirical phenomena. For example, the beneficial effect of chunking on recall often depends on familiarity with the chunks, as in the examples given previously. It is easily overlooked that the familiarity of the materials themselves is also important. For example, familiar words are recalled much better than nonwords ( Hulme et al., 1991 ) suggesting that words act as specialized phonological/semantic “chunks.” Similarly, grammatical sentences are recalled better than arbitrarily ordered lists or jumbled sentences ( Brener, 1940 ). The word–nonword and sentence superiority effects show that well-learned constraints on serial order (whether through syntax or phonotactics) can benefit recall. A related phenomenon, the Hebb repetition effect ( Hebb, 1961 ), can be seen in the laboratory: immediate serial recall for a specific random list gradually improves over successive trials when it becomes more familiar through being repeatedly but covertly presented interleaved among other lists.

The Importance of Working Memory

The laboratory tasks and benchmark findings outlined in the section “ Empirical Investigation and Key Findings ” have established its key characteristics, but the practical significance of working memory extends well beyond these phenomena into everyday cognition and learning. Notably the limits of working memory constrain what we can think about on a moment-to-moment basis and hence how quickly we can learn and what we can ultimately understand. An appreciation of the impact of working memory and its limitations is thus vitally important in the context of education (see e.g., Alloway & Gathercole, 2006 for a review). For example, individual differences in the capacity of phonological storage in verbal working memory are reciprocally linked to vocabulary acquisition in early childhood; children’s ability to repeat nonwords at age four (i.e., unfamiliar phonological sequences) predicts their vocabulary a year later. In turn, the emergence of vocabulary (i.e., phonological chunks) is associated with later improvements in nonword repetition ( Gathercole et al., 1992 ). It is not hard to imagine that this process amplifies the initial effect of variation in capacity, affecting literacy and then more advanced learning (potentially well beyond language abilities) that depends on reading. Working memory can similarly exert an influence on the emergence of numeracy and through it more advanced skills in arithmetic and mathematics. For example, kindergartners’ performance on a backward digit span task predicts their scores on a mathematics test a year later ( Gersten et al., 2005 ). In addition to these effects on the acquisition of foundational skills such as literacy and numeracy, working memory is important in maintaining and manipulating the information needed to carry out complex tasks in the classroom. Thus, students with lower working memory capacity can have difficulty retaining and following instructions ( Gathercole et al., 2008 ) again potentially hampering their ability to build more advanced skills and knowledge. Because of its critical involvement in classroom learning, working memory plays a central role in Cognitive Load Theory” ( Sweller, 2011 ) an influential educational framework which aims to incorporate principles derived from the architecture of human cognition into teaching methods.

Many measures of short-term memory and working memory show marked year-on-year improvement in childhood, with developmental change likely reflecting the maturation of several components that underpin performance ( Gathercole, 1999 ; Gathercole et al., 2004 ). These include changes in processes such as verbal recoding, subvocal rehearsal, the activation of temporary information and executive attentional control ( Camos & Barrouillet, 2011 ; Cowan et al., 2002 ; Hitch & Halliday, 1983 ). As might be expected given the centrality of working memory in the acquisition of language and numeracy, developmental disorders are commonly associated with reduced short-term or working memory capacity. Prominent examples include dyslexia ( Berninger et al., 2008 ), developmental language disorder ( Archibald & Gathercole, 2006 ; Montgomery et al., 2010 ), and dyscalculia ( Fias et al., 2013 ; McLean & Hitch, 1999 ). However, the nature of any causal role for working memory in developmental disorders has been controversial (see e.g., Masoura, 2006 ).

In adulthood, working memory capacity continues to limit the bandwidth that is available for cognitive operations, for example affecting planning and decision-making ( Gilhooly, 2005 ; Hinson et al., 2003 ). As we grow older, working memory capacity tends to decline, and there are some indications that this is associated with failing attention and greater vulnerability to distraction ( Hasher & Zacks, 1988 ; McNab et al., 2015 ; Park & Payer, 2006 ) rather than a mere reversal of earlier developmental gains. Across the entire lifespan, as it waxes and wanes, working memory plays an important part in shaping our daily experience.

Given its central role in constraining human cognitive abilities, extensive efforts have been made to develop interventions that can improve working memory, for example through computerized training programs. However, these efforts have so far met with limited success. Some working memory tasks show improvements with practice, but these effects tend to reflect near or intermediate transfer , specific to the trained task or (often closely-related) direct measures of working memory, rather than far transfer extending to more general improvements in other tasks thought to depend on working memory, such as reading comprehension or arithmetic ( Melby-Lervåg et al., 2016 ; Owen et al., 2010 ; Sala & Gobet, 2017 ). It has been argued that near and intermediate transfer effects arise through improvements in task-specific efficiency via refinement of strategies and long-term memory support (e.g., chunking) whereas more general benefits and far transfer would be expected to depend on the underlying capacity of attentional and storage systems ( von Bastian & Oberauer, 2014 ). The absence of clear evidence for far transfer despite such extensive research thus suggests that working memory capacity limits are a fundamental and unalterable feature of the human cognitive system.

Although it is perhaps premature to rule out the possibility of interventions that achieve increased working memory capacity, it appears at present that it can only be extended in specific contexts through more specialized training with particular tasks and materials. Paradoxically, this resistance to more general training may be what makes working memory so important; to the extent that its capacity limits are unavoidable, working memory helps to determine the scope of human cognition and spurs us to find strategies, technologies and cultural tools that allow us to go beyond them.

In conclusion, through the development of a powerful toolkit of experimental methods and of replicable empirical phenomena, the study of working memory function has provided many useful insights into interactions between attention and short-term memory. On the one hand these interactions can be used strategically to enhance goal-directed behavior and long-term learning while on the other they provide fundamental limits on cognition across the lifespan. Ongoing controversy over the structure of working memory relates to the difficulty in isolating these interactions from other facets of cognition, but there is little doubt about their importance in governing what we can and cannot do.

• Alloway, T. P. , & Gathercole, S. E. (2006). How does working memory work in the classroom? Education Research and Reviews , 1 (4), 134–139.
• Archibald, L. M. D. , & Gathercole, S. E. (2006). Short-term and working memory in specific language impairment . International Journal of Language & Communication Disorders , 41 (6), 675–693.
• Atkinson, R. C. , & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory (Vol. 2, pp. 89–195). Academic Press.
• Baars, B. J. (2005). Global workspace theory of consciousness: Toward a cognitive neuroscience of human experience . In S. Laureys (Ed.), Progress in brain research (Vol. 150, pp. 45–53). Elsevier.
• Baddeley, A. (1986). Working memory (pp. xi, 289). Clarendon Press.
• Baddeley, A. (2000). The episodic buffer: A new component of working memory ? Trends in Cognitive Sciences , 4 (11), 417–423.
• Baddeley, A. D. , & Hitch, G. (1974). Working memory . In G. H. Bower (Ed.), Psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press.
• Baddeley, A. D. , & Hitch, G. (1993). The recency effect: Implicit learning with explicit retrieval? Memory & Cognition , 21 (2), 146–155.
• Baddeley, A. , Hitch, G. , & Allen, R. (2021). A multicomponent model of working memory . In R. Logie , V. Camos , & N. Cowan (Eds.), Working memory: The state of the science . Oxford University Press.
• Baddeley, A. , & Logie, R. (1999). Working memory: The multiple-component model . In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 28–61). Cambridge University Press.
• Barnard, P. (1985). Interacting cognitive subsystems: A psycholinguistic approach to short-term memory. In A. W. Ellis (Ed.), Progress in the psychology of language (Vol. 2, pp. 197–258).
• Barrouillet, P. , & Camos, V. (2004). Time constraints and resource sharing in adults’ working memory spans. Journal of Experimental Psychology: General , 133 , 83–100.
• Barrouillet, P. , & Camos, V. (2021). The time-based resource-sharing model of working memory . In R. Logie , V. Camos , & N. Cowan (Eds.), Working memory . Oxford University Press.
• Bays, P. M. , & Husain, M. (2008). Dynamic shifts of limited working memory resources in human vision . Science , 321 (5890), 851–854.
• Berninger, V. W. , Raskind, W. , Richards, T. , Abbott, R. , & Stock, P. (2008). A multidisciplinary approach to understanding developmental dyslexia within working-memory architecture: Genotypes, phenotypes, brain, and instruction . Developmental Neuropsychology , 33 (6), 707–744.
• Brener, R. (1940). An experimental investigation of memory span . Journal of Experimental Psychology , 26 (5), 467–482.
• Camos, V. , & Barrouillet, P. (2011). Developmental change in working memory strategies: From passive maintenance to active refreshing . Developmental Psychology , 47 (3), 898–904.
• Camos, V. , Johnson, M. , Loaiza, V. , Portrat, S. , Souza, A. , & Vergauwe, E. (2018). What is attentional refreshing in working memory? Annals of the New York Academy of Sciences , 1424 (1), 19–32.
• Camos, V. , Lagner, P. , & Barrouillet, P. (2009). Two maintenance mechanisms of verbal information in working memory . Journal of Memory and Language , 61 (3), 457–469.
• Carruthers, P. (2017). The centered mind: What the science of working memory shows us about the nature of human thought (Reprint ed.). Oxford University Press.
• Cohen, N. J. , & Squire, L. R. (1980). Preserved learning and retention of pattern-analyzing skill in amnesia: Dissociation of knowing how and knowing that . Science , 210 (4466), 207–210.
• Corsi, P. (1972). Memory and the medial temporal region of the brain . McGill University.
• Cowan, N. (1999). An Embedded-Processes Model of working memory . In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 62–101). Cambridge University Press.
• Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity . Behavioral and Brain Sciences , 24 (1), 87–114.
• Cowan, N. , Morey, C. C. , & Naveh-Benjamin, M. (2021). An embedded-processes approach to working memory: How is it distinct from other approaches, and to what ends ? In R. Logie , V. Camos , & N. Cowan (Eds.), Working memory . Oxford University Press.
• Cowan, N. , Saults, J. S. , & Elliott, E. M. (2002). The search for what is fundamental in the development of working memory . In R. V. Kail & H. W. Reese (Eds.), Advances in child development and behavior (Vol. 29, pp. 1–49). JAI.
• Crannell, C. W. , & Parrish, J. M. (1957). A comparison of immediate memory span for digits, letters, and words . The Journal of Psychology: Interdisciplinary and Applied , 44 , 319–327.
• Daneman, M. , & Carpenter, P. A. (1980). Individual differences in working memory and reading . Journal of Verbal Learning and Verbal Behavior , 19 (4), 450–466.
• Engle, R. W. (2002). Working memory capacity as executive attention . Current Directions in Psychological Science , 11 (1), 19–23.
• Fias, W. , Menon, V. , & Szucs, D. (2013). Multiple components of developmental dyscalculia . Trends in Neuroscience and Education , 2 (2), 43–47.
• Fuchs, A. H. (1969). Recall for order and content of serial word lists in short-term memory . Journal of Experimental Psychology , 82 (1, Pt.1), 14–21.
• Gathercole, S. E. (1999). Cognitive approaches to the development of short-term memory . Trends in Cognitive Sciences , 3 (11), 410–419.
• Gathercole, S. E. , Durling, E. , Evans, M. , Jeffcock, S. , & Stone, S. (2008). Working memory abilities and children’s performance in laboratory analogues of classroom activities . Applied Cognitive Psychology , 22 (8), 1019–1037.
• Gathercole, S. E. , Pickering, S. J. , Ambridge, B. , & Wearing, H. (2004). The structure of working memory from 4 to 15 years of age . Developmental Psychology , 40 (2), 177–190.
• Gathercole, S. E. , Willis, C. S. , Emslie, H. , & Baddeley, A. D. (1992). Phonological memory and vocabulary development during the early school years: A longitudinal study . Developmental Psychology , 28 (5), 887–898.
• Gersten, R. , Jordan, N. C. , & Flojo, J. R. (2005). Early identification and interventions for students with mathematics difficulties . Journal of Learning Disabilities , 38 (4), 293–304.
• Gilhooly, K. J. (2005). Working memory and planning. The Cognitive Psychology of Planning , 71–88.
• Gomez-Lavin, J. (2021). Working memory is not a natural kind and cannot explain central cognition . Review of Philosophy and Psychology , 12 (2), 199–225.
• Hasher, L. , & Zacks, R. T. (1988). Working memory, comprehension, and aging: A review and a new view . In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 22, pp. 193–225). Academic Press.
• Hassin, R. R. , Bargh, J. A. , Engell, A. D. , & McCulloch, K. C. (2009). Implicit working memory . Consciousness and Cognition , 18 (3), 665–678.
• Hebb, D. O. (1961). Distinctive features of learning in the higher animal. In J. F. Delafresnaye (Ed.), Brain mechanisms and learning (pp. 37–46). Blackwell.
• Hinson, J. M. , Jameson, T. L. , & Whitney, P. (2003). Impulsive decision making and working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition , 29 (2), 298.
• Hitch, G. J. , Allen, R. J. , & Baddeley, A. D. (2020). Attention and binding in visual working memory: Two forms of attention and two kinds of buffer storage. Attention, Perception, & Psychophysics , 82 (1), 280–293.
• Hitch, G. J. , & Halliday, M. S. (1983). Working memory in children . Philosophical Transactions of the Royal Society of London. B, Biological Sciences , 302 (1110), 325–340.
• Hulme, C. , Maughan, S. , & Brown, G. D. A. (1991). Memory for familiar and unfamiliar words: Evidence for a long-term memory contribution to short-term memory span . Journal of Memory and Language , 30 (6), 685–701.
• Jahnke, J. C. (1963). Serial position effects in immediate serial recall . Journal of Verbal Learning and Verbal Behavior , 2 (3), 284–287.
• Jalbert, A. , Saint-Aubin, J. , & Tremblay, S. (2008). Short article: Visual similarity in short-term recall for where and when . Quarterly Journal of Experimental Psychology , 61 (3), 353–360.
• Kane, M. J. , Conway, A. R. , Miura, T. K. , & Colflesh, G. J. (2007). Working memory, attention control, and the N-back task: A question of construct validity. Journal of Experimental Psychology: Learning, Memory, and Cognition , 33 (3), 615.
• Kessels, R. P. C. , van den Berg, E. , Ruis, C. , & Brands, A. M. A. (2008). The backward span of the Corsi Block-Tapping Task and its association with the WAIS-III Digit Span . Assessment , 15 (4), 426–434.
• Logie R.H. (1995). Visuo-Spatial Working Memory . Psychology Press.
• Logie, R. H. (1996). The seven ages of working memory. In J. T. E. Richardson , R. W. Engle , L. Hasher , R. H. Logie , E. R. Stoltzfus , & R. T. Zacks (Eds.), Working memory and human cognition (pp. 31–65).
• Logie, R. , Camos, V. , & Cowan, N. (Eds.). (2021). Working memory: The state of the science (1st ed.). Oxford University Press.
• Luck, S. J. , & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions . Nature , 390 (6657), 279–281.
• Ma, W. J. , Husain, M. , & Bays, P. M. (2014). Changing concepts of working memory . Nature Neuroscience , 17 (3), 347–356.
• Martin, C. (1993). Short-term memory and sentence processing: Evidence from neuropsychology. Memory & Cognition , 21 (2), 176–183.
• Masoura, E. V. (2006). Establishing the link between working memory function and learning disabilities. Learning Disabilities: A Contemporary Journal , 4 (2), 29–41.
• McElree, B. , & Dosher, B. A. (1989). Serial position and set size in short-term memory: The time course of recognition . Journal of Experimental Psychology: General , 118 (4), 346–373.
• McLean, J. F. , & Hitch, G. J. (1999). Working memory impairments in children with specific arithmetic learning difficulties . Journal of Experimental Child Psychology , 74 (3), 240–260.
• McNab, F. , Zeidman, P. , Rutledge, R. B. , Smittenaar, P. , Brown, H. R. , Adams, R. A. , & Dolan, R. J. (2015). Age-related changes in working memory and the ability to ignore distraction . Proceedings of the National Academy of Sciences , 112 (20), 6515–6518.
• Melby-Lervåg, M. , Redick, T. S. , & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence or other measures of “far transfer”: Evidence From a meta-analytic review . Perspectives on Psychological Science , 11 (4), 512–534.
• Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information . Psychological Review , 63 (2), 81–97.
• Milner, B. (1971). Interhemispheric differences in the localization of psychological processes in man . British Medical Bulletin , 27 (3), 272–277.
• Miyake, A. , & Shah, P. (Eds.). (1999). Models of working memory: Mechanisms of active maintenance and executive control . Cambridge University Press.
• Montgomery, J. W. , Magimairaj, B. M. , & Finney, M. C. (2010). Working memory and specific language impairment: An update on the relation and perspectives on assessment and treatment . American Journal of Speech-Language Pathology , 19 (1), 78–94.
• Morris, N. , & Jones, D. M. (1990). Memory updating in working memory: The role of the central executive . British Journal of Psychology , 81 (2), 111–121.
• Murray, D. J. (1967). The role of speech responses in short-term memory . Canadian Journal of Psychology/Revue Canadienne de Psychologie , 21 (3), 263.
• Myers, N. E. , Stokes, M. G. , & Nobre, A. C. (2017). Prioritizing information during working memory: Beyond sustained internal attention . Trends in Cognitive Sciences , 21 (6), 449–461.
• Oberauer, K. (2002). Access to information in working memory: Exploring the focus of attention. Journal of Experimental Psychology: Learning, Memory, and Cognition , 28 , 411–421.
• Oberauer, K. (2021). Towards a theory of working memory: From metaphors to mechanisms . In R. Logie , V. Camos , & N. Cowan (Eds.), Working memory . Oxford University Press.
• Oberauer, K. , Lewandowsky, S. , Awh, E. , Brown, G. D. A. , Conway, A. , Cowan, N. , Donkin, C. , Farrell, S. , Hitch, G. J. , Hurlstone, M. J. , Ma, W. J. , Morey, C. C. , Nee, D. E. , Schweppe, J. , Vergauwe, E. , & Ward, G. (2018). Benchmarks for models of short-term and working memory . Psychological Bulletin , 144 (9), 885–958.
• Owen, A. M. , Hampshire, A. , Grahn, J. A. , Stenton, R. , Dajani, S. , Burns, A. S. , Howard, R. J. , & Ballard, C. G. (2010). Putting brain training to the test . Nature , 465 (7299), 775–778.
• Owen, A. M. , McMillan, K. M. , Laird, A. R. , & Bullmore, E. (2005). N-back working memory paradigm: A meta-analysis of normative functional neuroimaging studies . Human Brain Mapping , 25 (1), 46–59.
• Park, D. C. , & Payer, D. (2006). Working memory across the adult lifespan . In Lifespan cognition: Mechanisms of change (pp. 128–142). Oxford University Press.
• Persuh, M. , LaRock, E. , & Berger, J. (2018). Working memory and consciousness: The current state of play . Frontiers in Human Neuroscience , 12 .
• Pesenti, M. , Seron, X. , Samson, D. , & Duroux, B. (1999). Basic and exceptional calculation abilities in a calculating prodigy: A case study . Mathematical Cognition , 5 (2), 97–148.
• Peterson, L. , & Peterson, M. J. (1959). Short-term retention of individual verbal items . Journal of Experimental Psychology , 58 (3), 193–198.
• Posner, M. I. , & Konick, A. F. (1966). On the role of interference in short-term retention . Journal of Experimental Psychology , 72 (2), 221–231.
• Postle, B. R. (2006). Working memory as an emergent property of the mind and brain . Neuroscience , 139 (1), 23–38.
• Postman, L. , & Phillips, L. W. (1965). Short-term temporal changes in free recall . Quarterly Journal of Experimental Psychology , 17 (2), 132–138.
• Raghubar, K. P. , Barnes, M. A. , & Hecht, S. A. (2010). Working memory and mathematics: A review of developmental, individual difference, and cognitive approaches . Learning and Individual Differences , 20 (2), 110–122.
• Ryan, J. (1969). Grouping and short-term memory: Different means and patterns of grouping . Quarterly Journal of Experimental Psychology , 21 (2), 137–147.
• Sala, G. , & Gobet, F. (2017). Working memory training in typically developing children: A meta-analysis of the available evidence . Developmental Psychology , 53 (4), 671–685.
• Soto, D. , Mäntylä, T. , & Silvanto, J. (2011). Working memory without consciousness . Current Biology , 21 (22), R912–R913.
• Stigler, J. W. (1984). “Mental abacus”: The effect of abacus training on Chinese children’s mental calculation . Cognitive Psychology , 16 (2), 145–176.
• Sweller, J. (2011). Cognitive load theory . In J. P. Mestre & B. H. Ross (Eds.), Psychology of learning and motivation (Vol. 55, pp. 37–76). Academic Press.
• Vandierendonck, A. (2021). Multicomponent working memory system with distributed executive control . In R. Logie , V. Camos , & N. Cowan (Eds.), Working memory . Oxford University Press.
• von Bastian, C. C. , & Oberauer, K. (2014). Effects and mechanisms of working memory training: A review . Psychological Research , 78 (6), 803–820.

1. However, note that, in at least one study ( Kane et al., 2007 ) n-back performance correlated only weakly with a measure of span, suggesting that, despite face validity, it may tax distinct cognitive resources.

Related Articles

• Working Memory and Cognitive Aging
• Memory Rehabilitation in Healthy Aging
• Culture and memory

Printed from Oxford Research Encyclopedias, Psychology. Under the terms of the licence agreement, an individual user may print out a single article for personal use (for details see Privacy Policy and Legal Notice).

date: 13 September 2024

• Cookie Policy
• Privacy Policy
• Legal Notice
• Accessibility
• [195.158.225.244]
• 195.158.225.244

Character limit 500 /500

Our systems are now restored following recent technical disruption, and we’re working hard to catch up on publishing. We apologise for the inconvenience caused. Find out more: https://www.cambridge.org/universitypress/about-us/news-and-blogs/cambridge-university-press-publishing-update-following-technical-disruption

We use cookies to distinguish you from other users and to provide you with a better experience on our websites. Close this message to accept cookies or find out how to manage your cookie settings .

Login Alert

• > Journals
• > The Educational and Developmental Psychologist
• > Volume 30 Issue 2
• > Working Memory: The What, the Why, and the How

Article contents

What is working memory an introduction, why is working memory important, how can we diagnose and support working memory, cognitive tests of working memory, working memory: the what, the why, and the how.

Published online by Cambridge University Press:  13 November 2013

Working memory, our ability to work with information, plays an important role in learning from kindergarten to the college years. In this article, we review the what, the why, and the how of working memory. First, we explore the relationship between working memory, short-term memory, and long-term memory. We also investigate research on the link between whether environmental factors, such as financial background and mother's educational level, affect working memory. In the next section — the why of working memory — we compare the predictive nature of working memory and IQ in learning outcomes. While IQ typically measures the knowledge acquired by the student, working memory measures what they do with that knowledge. Working memory skills are linked to key learning outcomes, including reading and math. In the final section, we present classroom strategies to support working memory. We also review current research on the efficacy of working memory training.

The aim of this review is to introduce a cognitive skill that has been linked with learning — working memory. We discuss the relationship between working memory and other related cognitive skills, such as short-term memory and long-term memory. Next, we introduce research on the role of working memory in learning and compare it with verbal and nonverbal IQ skills. We conclude by providing classroom strategies that educators can adopt to support working memory. We also discuss research on the efficacy of working memory training.

Working memory is our ability to work with information (Alloway, Reference Alloway 2010 ). This higher-level skill is involved in directing attention to a task despite distraction or interference (Cowan, Reference Cowan 2006 ; Engle, Tuholski, Laughlin, & Conway, Reference Engle, Tuholski, Laughlin and Conway 1999 ). Working memory is linked to a range of cognitive activities during the school years, from reasoning tasks to verbal comprehension to mathematical skills (see Cowan & Alloway, Reference Cowan, Alloway, Courage and Cowan 2008 , for a review).

Working memory is distinct from short-term memory, which typically refers to remembering information for a brief period, usually a few seconds (Alloway, Gathercole, & Pickering, Reference Alloway, Gathercole and Pickering 2006 ; see McGrew, Reference McGrew 2009 ). We utilise short-term memory when we remember someone's name, or a phone number, or a title of a book. Typically, this information will be forgotten if it is not rehearsed. Imagine that you are driving to a new school for a meeting. You lose your way and stop at a store to ask for directions. You may repeat the information to yourself over and over again as you walk back to your car so you do not forget. At this point, you are using your short-term memory to remember the directions. Now you get back inside your car and start driving. As you recite the directions to yourself, you look around and match them to the road names. Is this where you make that right turn? Where do you make that second left? Now you are using your working memory as you are applying or using the information that you were given. It is much the same in the classroom. When you give a student a set of instructions, they use their short-term memory to repeat it to themselves. However, by the time they get back to their desk and have to carry out the first task in the set of instructions, chances are if they have poor working memory, they will have forgotten what to do. The process of repeating the information and then carrying out the individual steps relies on working memory.

Working memory is also distinct from long-term memory, though there is close relationship between them, much like a two-way street. Long-term memory refers to memories from our childhood, but it also refers to the knowledge that we have accumulated over the years, such as facts about a country, mathematical knowledge, and grammar rules. One goal of working memory is to transfer new information to our long-term memory. For example, if we are planning a trip to a country that we have not visited before, we use our working memory to retain and transfer the knowledge we learn about that country to our long-term memory. In turn, we can draw on long-term memory to form associations between a familiar place and the new country we are about to visit.

Working Memory and the Environment

Given that working memory is linked to long-term knowledge stores, one issue is the extent to which working memory is influenced by environmental factors that contribute to knowledge acquisition. Of particular interest is the impact of socio-economic status (SES) as this has long been linked with school success, and the income-achievement gap is evident in kindergarten and accelerates over time (Kaplan et al., Reference Kaplan, Turrell, Lynch, Everson, Helkela and Salonen 2001 ). However, with respect to working memory, SES does not appear to have a significant impact on performance. For example, children from low-SES areas in South America did not differ significantly from their middle-SES peers in some working memory tests, although their vocabulary scores, reflecting knowledge-based skills, were considerably worse (Engel, Heloisa Dos Santos, & Gathercole, Reference Engel, Heloisa Dos Santos and Gathercole 2008 ). Dutch studies investigating differences between immigrant children that typically reside in low-income areas (indexed by parental educational), and comparatively wealthier native language speakers found that the former group performed at the same level as their native language peers in working memory tests when tested in their own language (Messer, Leseman, Mayo, & Boom, Reference Messer, Leseman, Mayo and Boom 2010 ; also Leseman, Scheele, Mayo, & Messer, Reference Leseman, Scheele, Mayo and Messer 2007 ). Typically, there are multiple indices of SES, such as family income and parental education, though studies of only parental education have reported that SES is not linked to working memory performance (Alloway, Gathercole, Willis, & Adams, Reference Alloway, Gathercole, Willis and Adams 2004 ; Messer et al., Reference Messer, Leseman, Mayo and Boom 2010 ).

However, some researchers have reported a differentiation of working memory performance as a function of SES levels (Noble, McCandliss, & Farah, Reference Noble, McCandliss and Farah 2007 ; Noble, Norman, & Farah, Reference Noble, Norman and Farah 2005 ). There are two explanations for this disparity in the impact of SES on working memory. The first explanation is sample age, as the chronic stress hypothesis suggests that the prolonged exposure to poverty can result in chronic stress, which in turn leads to reductions in working memory performance (Evans & Schamberg, Reference Evans and Schamberg 2009 ). According to this hypothesis, SES would have a greater impact on older children compared to younger ones. This pattern appears to hold true as studies with older samples (10–13 years) found differences in spatial memory (keeping a location in mind) and the n -back task (Farah et al., Reference Farah, Shera, Savage, Betancourt, Giannetta, Brodsky, Malmud and Hurt 2006 ). Evans and Schamberg ( Reference Evans and Schamberg 2009 ) also reported working memory deficits in their 17-year-olds, as a function of SES. In contrast, studies with young populations reported that working memory was relatively unaffected by SES levels (e.g., Alloway et al., Reference Alloway, Gathercole, Willis and Adams 2004 , with British 4- to 5-year-olds; Engel et al., Reference Engel, Heloisa Dos Santos and Gathercole 2008 , with Brazilian 6- to 7-year-olds; Messer et al., Reference Messer, Leseman, Mayo and Boom 2010 , with Dutch 4-year-olds). One exception is a study with first-graders in New York City (Noble et al., Reference Noble, McCandliss and Farah 2007 ; Noble et al., Reference Noble, Norman and Farah 2005 ), which will be discussed further in the following paragraph.

The tasks used to measure working memory can also affect findings as some studies use tasks that are similar to short-term memory ones (e.g., Noble et al., Reference Noble, McCandliss and Farah 2007 ). Such tasks do not involve any processing of information, which is reflective of working memory, and may rely more on knowledge structures (Alloway et al., Reference Alloway, Gathercole and Pickering 2006 ). As a result, performance may be more sensitive to SES variations, which may account for Noble et al.'s findings. Indeed, Farah et al. ( Reference Farah, Shera, Savage, Betancourt, Giannetta, Brodsky, Malmud and Hurt 2006 ) argued that SES has differential effects on tasks associated with different neurocognitive systems. They reported the greatest effects in the left perisylvian/language system (measured by tests of receptive vocabulary and grammar), but no effect on the parietal lobe/spatial cognition (measured by mental rotation tasks).

Working memory is critical for a variety of activities at school, from complex subjects such as reading comprehension, mental arithmetic, and word problems to simple tasks like copying from the board and navigating around school. It is also important from kindergarten (Alloway et al., Reference Alloway, Gathercole, Adams, Willis, Eaglen and Lamont 2005 ) to the tertiary level (Alloway & Gregory, Reference Alloway, Doherty-Sneddon and Forbes 2012 ).

Working Memory and Reading

A key foundational skill in reading success is known as phonological awareness , where the child must dissect a word into its parts, such as rhyming words or words with the same initial sounds, and the ability to name pictures rapidly. A 5-year longitudinal study of several hundred children who were tracked from kindergarten through fourth grade confirmed that phonological awareness skills predicted reading proficiency (Wagner et al., Reference Wagner, Torgesen, Rashotte, Hecht, Barker, Burgess, Donahue and Garon 1997 ).

Working memory is also highly predictive of reading success. In typically developing children, scores on working memory tasks predict reading achievement independently of measures of phonological awareness (Swanson & Beebe-Frankenberger, Reference Swanson and Beebe-Frankenberger 2004 ). One explanation for why working memory is so critical for reading is that we use our ‘Post-it Note’ (or working memory capacity) to keep all the relevant speech sounds in mind, match them up with the corresponding letters, and the combine them to read the words. Indeed, children with reading difficulties have been found to have a limited capacity for processing and storing information (De Jong, Reference De Jong 1998 ), and often show significant and marked decrements on working memory tasks (Siegel & Ryan, Reference Siegel and Ryan 1989 ).

Working Memory and Mathematics

Working memory is also linked to math outcomes: low working memory scores are closely related to poor performance on arithmetic word problems (Swanson & Sachse-Lee, Reference Swanson and Sachse-Lee 2001 ; also Alloway & Passolunghi, Reference Alloway and Passolunghi 2011 ) and poor computational skills (Bull & Scerif, Reference Bull and Scerif 2000 ; Geary, Hoard, & Hamson, Reference Geary, Hoard and Hamson 1999 ).

Although there is also a close relationship between mathematical skills and working memory, this is mediated by the age of the child, as well as the task. Verbal working memory plays a strong role in math skills in 7-year-olds (Bull & Scerif, Reference Bull and Scerif 2001 ) and is also a reliable indicator of mathematical difficulties in the first year of formal schooling (Gersten, Jordan, & Flojo, Reference Gersten, Jordan and Flojo 2005 ). However, once children reach adolescence, working memory is no longer significantly linked to mathematical skills (Reuhkala, Reference Reuhkala 2001 ). One explanation for this change is that verbal working memory plays a crucial role for basic arithmetic (both to learn arithmetic facts and to retain relevant data such as carried digits) but that as children get older other factors, such as number knowledge and strategies, play a greater role (Thevenot & Oakhill, Reference Thevenot and Oakhill 2005 ). Low working memory scores are related to poor computational skills (Bull & Scerif, Reference Bull and Scerif 2001 ; Geary, Hoard, & Hamson, Reference Geary, Hoard and Hamson 1999 ) and poor performance on arithmetic word problems (Swanson & Sachse-Lee, Reference Swanson and Sachse-Lee 2001 ).

Visuo-spatial memory is also closely linked with mathematical skills. It has been suggested that visuo-spatial memory functions as a mental blackboard, supporting number representation, such as place value and alignment in columns, in counting and arithmetic (D'Amico & Guarnera, Reference D'Amico and Guarnera 2005 ; Geary, Reference Geary 1990 ; McLean & Hitch, Reference McLean and Hitch 1999 ). Specific associations have been found between visuo-spatial memory and encoding in problems presented visually (Logie et al., Reference Logie, Gilhooly and Wynn 1994 ), and in multi-digit operations (Heathcote, Reference Heathcote 1994 ). Visuo-spatial memory skills also uniquely predict performance in nonverbal problems, such as sums presented with blocks, in pre-school children (Rasmussen & Bisanz, Reference Rasmussen and Bisanz 2005 ).

Working Memory Versus IQ Footnote 1

Many studies have demonstrated that both IQ and working memory are related to learning. In the research lab, we investigated which is more important. This issue is important so that educators can target and support the cognitive skills that underpin success in learning. In order to investigate how well IQ and working memory would predict reading, writing, and math skills, a group of 5-year-olds ( n = 194) was tested as they started kindergarten and tracked over a 6-year period. The findings at the first time point (age 5) indicated that working memory was a significant predictor of reading, writing and math. Children with high working memory did well in reading, writing, and math; while those with low working memory struggled in these tasks (Alloway et al., Reference Alloway, Gathercole, Adams, Willis, Eaglen and Lamont 2005 ).

The children were tested again when they were 11 years old in order to explore the best predictors of learning outcomes over time — working memory or Verbal IQ/Performance IQ (VIQ/PIQ). They were also tested on standardised tests of language and math. The results indicated that a student's working memory ability at 5 years of age was a significant predictor of language and math scores 6 years later (Alloway & Alloway, Reference Alloway, Gathercole and Elliott 2010 ). This finding is important as it indicates that while IQ is still viewed as a benchmark of success, other skills, such as working memory, may provide more useful information on a student's potential to learn.

Numerous studies have demonstrated that working memory is a distinct skill from IQ (Cain, Oakhill, & Bryant, Reference Cain, Oakhill and Bryant 2004 ; Siegel, Reference Siegel 1988 ), and uniquely predicts learning outcomes. For example, working memory skills predict a child's performance in language and math, even after a child's IQ scores have been statistically accounted (Gathercole, Alloway, Willis & Adams, Reference Gathercole, Alloway, Willis and Adams 2006 ; Nation, Adams, Bowyer-Crane, & Snowling, Reference Nation, Adams, Bowyer-Crane and Snowling 1999 ; Stothard & Hulme, Reference Stothard and Hulme 1992 ; for a review see Swanson & Saez, Reference Swanson, Saez, Swanson, Graham and Harris 2003 ). The importance of working memory in learning is not just limited to children. This same pattern is evidenced at the university level as well: working memory is a better predictor of grades than entrance exams like SAT scores (Engle et al., Reference Engle, Tuholski, Laughlin and Conway 1999 ).

Why is working memory a better predictor of learning than IQ? Working memory tests measure something different from IQ tests: working memory is an indicator of our potential to learn. A common working memory test is to remember a sequence of numbers in the reverse order that it was presented to you. If students struggle in this test, it is not because they do not know how to count, or understand number magnitude. It doesn't even matter whether they can recognise the numbers. If they struggle in this working memory test, it is often because their ‘Post-it Note’ (or working memory capacity) isn't big enough to remember three or four numbers. Working memory is an accurate predictor of learning from kindergarten to college because it measures students’ ability to learn, rather than what they have learned.

In contrast, other measurements like school tests and IQ tests measure knowledge that they have already learned. If students do well on one of these tests, it is because they know the information they are tested on. Likewise, many aspects of IQ tests also measure the knowledge that we have built up. A commonly used measure of IQ is a vocabulary test. If students know the definition of a word like ‘bicycle’ or ‘police’, then they will likely get a high IQ score. However, if they do not know the definitions of these words or perhaps do not articulate them well, this will be reflected in a low IQ score. In this way, IQ tests are very different from working memory tests because they measure how much students know and well they can articulate this knowledge.

One research project involved two different schools: one was in an urban, developed area, while the other was in an underprivileged neighborhood (Alloway, Alloway, & Wootan, Reference Alloway, Alloway and Wootan 2013 ). As part of the project, students’ IQ was tested using a vocabulary test. One of the vocabulary words — police — drew very different responses. Students from the urban school provided definitions relating to safety or uniforms, which corresponded to the examples in the manual. However, those from the underprivileged neighborhood responded with statements such as ‘I don't like police’ or ‘They are bad because they took my dad away’. Although both responses were drawn directly from the children's experiences, only one type of answer matched the IQ manual's definitions. This example illustrates how performance on IQ tests is strongly driven by a child's background and experiences.

Testing Working Memory

How can you detect working memory problems in a student? If we fall and break a leg, a cast is clearly visible. Yet working memory problems are often hidden from family, friends, and even teachers. In interviews with classroom teachers, we found that working memory failure in a student is often overlooked (Alloway, Doherty-Sneddon, & Forbes, Reference Alloway, Doherty-Sneddon and Forbes 2012 ). Instead, the student is often thought of as lazy or unmotivated. Comments such as ‘You are not trying hard enough’ or ‘Stop playing around and just focus’ are often directed towards the student with working memory problems.

Jenny, 14 years, had difficulty staying on task and always seemed to be two steps behind in her class assignments. In Science class, she had to label and remember the planets in the solar system. The next step was to apply this information to an in-class project. However, when the researcher (EC) walked over to her desk, he noticed that she was still working on the labels — an activity that should have been completed the previous day.

She displayed similar behaviour in her English class. When her teacher asked her to express her thoughts on an essay that was just read to the class, she answered with remarks pertaining to the essay read the day before. When the teacher reprimanded her for not paying attention, she seemed confused and did not understand what she had done. Her mind always seemed to be on the previous day's work, and her performance in classes suffered as a result. She was in a vicious cycle of being a day behind because she could not maintain focus long enough to complete any assignments. Not only were her grades suffering, but also she was frequently frustrated due to the constant reprimands and poor performance she dealt with on a regular basis.

Jenny is an example of classroom behaviour that is characteristic of a student with working memory difficulties. It is not uncommon for working memory difficulties to be regarded as attention problems. Students can lack direction, appear unmotivated, or simply disinterested in the activity. The Working Memory Rating Scale (WMRS) is a behavioural rating scale developed for educators to help them easily identify students with working memory deficits. It consists of 20 descriptions of behaviours characteristic of children with working memory deficits. Teachers rate how typical each behavior was of a particular child, using a 4-point scale ranging from (0) not typical at all to (1) occasionally to (2) fairly typical to (3) very typical .

A starting point in developing the items in the WMRS was an observational study of students with low working memory but typical scores in IQ tests. Compared with classmates with average working memory, the low memory students frequently forgot instructions, struggled to cope with tasks involving simultaneous processing and storage, and lost track of their place in complex tasks. The most common consequence of these failures was that the students abandoned the activity without completing it.

As the WMRS focuses solely on working memory–related problems in a single scale, it does not require any training in psychometric assessment prior to use. It is valuable not only as a diagnostic screening tool for identifying children at risk of poor working memory, but also in illustrating both the classroom situations in which working memory failures frequently arise, and the profile of difficulties typically faced by students with working memory difficulties. The scores are normed for each age group, which means that they are representative of typical classroom behavior for each age group. One item in the WMRS is ‘needs regular reminders of each step in the written task’. The classroom teacher has to rate how typical this behaviour is of the student and compare their score to the manual. A 5-year-old would need more reminders than a 10-year-old, which is reflected in the scoring of the WMRS. The scoring is color-coded to make it easy to interpret. For example, a score in the Green range indicates that it is unlikely that the student has a working memory impairment. If a student's score falls in the Yellow range, it is possible that they have a working memory impairment and further assessment is recommended. Scores in the red range indicate the presence of a working memory and targeted support is recommended.

The WMRS has been validated against other behaviour rating scales, such as the Conners Teacher Rating Scale and the BRIEF (Alloway, Elliott, & Place, Reference Alloway and Alloway 2010 ). The WMRS measures behaviour that is different from that represented in these other rating scales, and thus reliably identifies students with poor working memory. The WMRS has also been compared to cognitive tests of working memory, IQ, and academic attainment. The majority of students considered by their teachers to have problematic behaviours (i.e., typical of poor working memory) are more likely to have low working memory scores and achieve low grades (Alloway, Gathercole, & Elliott, Reference Alloway, Elliott and Place 2010 ).

It is important to know that students who display poor working memory behavior will not necessarily have low IQ scores. Many of them can have average IQ scores. Yet, it is working memory overload that leads to all the behaviours we have discussed and their loss of focus in the task can appear to be inattentive and distracted to others. The WMRS enables teachers to use their knowledge of the student to produce an indicator of how likely it is that the child has a working memory problem. Thus, it provides a valuable first step in detecting possible working memory failures.

Mary, 14 years, struggled in writing assignments. If a writing assignment extended over several days, she had a difficult time remembering her train of thought from her previous writing session. She needed hands-on support from her teacher and asked questions frequently to guide her activities. When she was asked to read her writing out aloud, her reading was uncertain and sounded similar to reading an unfamiliar text. She would skip lines and mingle sentences together from different parts of the paper. She required extra attention and guidance in order to complete her assignments.

Educators are growing increasingly aware that students like Mary have working memory difficulties that can impact their learning. Test publishers are also recognising the importance of working memory in education. Many standardised IQ test batteries, such as the Wechsler's Intelligence Scale, Stanford-Binet, and Woodcock Johnson, all include working memory tests as part of their assessment. However, these batteries are limited as they do not include visuo-spatial working memory tests and so do not provide a working memory profile of strengths and difficulties that can inform individualised education plans.

In order to address this need, the Alloway Working Memory Assessment (AWMA; Alloway, Reference Alloway 2007 ) Alloway Working Memory Assessment — 2 (Alloway, Reference Alloway 2012B ) were developed. This standardised battery is fully automated and provides assessments of verbal and visuo-spatial working memory, making it convenient for teachers and educational professionals alike, to screen individuals for significant working memory problems. Working memory tasks involve both remembering and processing information, while short-term memory is assessed using tasks that involve only remembering information. The AWMA uses a variety of stimuli. For example, tests of verbal working memory consist of letters and numbers, and the visual-spatial working memory tests include dot locations and three-dimensional arrays of blocks. Multiple methods of assessing the same underlying aspect of working memory allow the test administrator to distinguish whether a student has a working memory deficit or as difficulty in processing a particular type of stimulus.

The AWMA uses a span procedure, which makes it particularly suitable when testing both children and adults. The number of items to be remembered is increased over successive trials until the individual begins to struggle. Memory span (capacity) is the maximum amount of information that an individual can remember accurately. For example, in the backward digit recall test (verbal working memory), an individual remembers numbers in backwards order. If they are unable to remember five numbers, then the task ends and the individual's memory capacity is four items.

1. The AWMA is quick: It takes 10 minutes

The Screener version of the AWMA can be used to screen students with suspected working memory difficulties. It consists of the following two tests: Processing Letter Recall and Mr X. For a more detailed memory profile of the student, the AWMA also includes a Short Form, which includes tests of verbal short-term memory and visual-spatial short-term memory in addition to the working memory ones in the Screener. The administration time is approximately 20 minutes for all individuals. A Long Form with multiple assessments of each memory component is also included in the AWMA. It takes approximately 30 minutes to administer. One reason that the AWMA is easy for teachers to use is that the program automatically generates a report with standard scores and percentiles that is easy to interpret once the test is completed.

2. The AWMA has good reliability

Test reliability refers to the consistency with which a test can accurately measure what it aims to do. If an individual's performance remains consistent over repeated trials, it is considered to be reliable. Thousands of students have been tested and then re-tested on the AWMA 6 weeks and 1 year apart. The test–retest scores for the AWMA are high, indicating that the AWMA captures the stability of working memory over time (Alloway, Gathercole, Kirkwood, & Elliott, Reference Alloway, Gathercole, Kirkwood and Elliott 2008 ).

3. The AWMA has good validity

Test validity refers to whether a test accurately measures the skills it is designed to measure. In order to establish test validity, I took a group of students with poor working memory as identified by the AWMA and compared their scores in the Wechsler Working Memory Index (WMI; Alloway, Gathercole, Kirkwood, & Elliott, Reference Alloway, Gathercole, Kirkwood and Elliott 2009 ). The majority of students who achieved low scores on the AWMA also scored poorly on the Wechsler WMI. This pattern establishes that AWMA provides a valid measure of working memory.

The AWMA is effective in identifying students at risk. In a study of 3,000 students, the majority of students with poor working memory also scored poorly on a standardised attainment test and vocabulary (Alloway et al., Reference Alloway, Gathercole, Kirkwood and Elliott 2009 ). Scores on the AWMA can also identify those who need extra support in the classroom, as well as those who take longer to process information and so would benefit from extra time during assessments.

Supporting Working Memory Through Strategies

Classroom teachers can make small tweaks in the daily routine of the student to support their learning.

1. Detect working memory failures

Is the student struggling to keep up with their peers? Are they beginning to disengage from the activity? Are they acting out in frustration? Once you have identified these signs in a student, you can follow the next two recommendations.

2. Break down information

If an activity exceeds the working memory capacity of a student, they will be unable to complete the task.

3. Build long-term knowledge

This process can foster automaticity of knowledge in the student, which can ease the likelihood of working memory overload.

The following case studies illustrate how these steps can be implemented in a classroom setting.

Jimmy, 10 years, had difficulty issues recalling information, as well as completing simple tasks. His writing skills were poor and his reading comprehension was also much lower than the rest of his classmates.

1. Detecting working memory failures . When the researcher (EC) reviewed his lesson plans, he realised that the time that the class worked on certain assignments changed every day. Jimmy found it difficult to work within this varying schedule and was frequently frustrated. In order to support his learning and keep him apace with his classmates, the researcher started by developing a structured schedule for him. For example, every day at the same time he would complete writing activities, regardless of what the rest of the class was doing. Now that his day was structured, he knew exactly what to anticipate and was mentally ready to tackle his next assignment. His behaviour improved as a result.

2. Break down information . During writing assignments, the researcher would break down complex sentences and have Jimmy write one sentence at a time. After each paragraph, the researcher would read it aloud to Jimmy and then ask him to read it. Eventually he was able to write multiple sentences at a time without prompting, and read the paragraph aloud to the researcher before it was read to him.

3. Build long-term knowledge . Each day, the researcher would review the multi-syllable words with Jimmy and reinforce meaning. The next day the researcher would ask him what the word meant. The word ‘because’ perplexed him at first. One day, they used it in a sentence and read it together, and the researcher explained the meaning of the word, as well as the proper usage. The next day during Jimmy's writing assignment, he had to use the word in one of his sentences. He was able to use the word effectively and continued to better its usage throughout the week. Developing a schedule, breaking down the writing assignments, and explaining the meaning of words allowed him to catch up with his classmates in writing and reading assignments.

Janine, 11 years, was having difficulty with manipulating three and four digit numbers.

1. Detecting working memory failures . When the researcher (EC) first assessed her ability, he realised she was proficient with two-digit multiplication, but the borrowing system with three digits confused her. The same issue was evident with her long division.

2. Break down information. The researcher began with asking her to add multi-digit numbers together (e.g., 345 + 678). After successfully completing this task, he asked her to multiply the same numbers together, guiding her through each step along the way and talking about each step. They spent a week doing this together. Her homework assignments had to be completed in a quiet room away from televisions and radios. If she had any difficulties with the math problems, she was instructed to make a note next to the problem and move onto the next one (to avoid her adopting incorrect techniques). The following day they would review her notes and the next homework assignment contained problems that mimicked the ones she had difficulties with. Eventually she was able to confidently perform complex multiplication with ease.

The next challenge was long division. The multiplication sessions began by going over simple division (e.g., 16/4) and going through them step by step. Janine began by only dividing numbers that were even, and eventually integrated simple numbers with decimals. After integrating decimals, she was moved on to even-number long division (e.g., 100/5). The researcher went through these step by step with her until she could do them with ease. She began to remember the steps in the process, as well as the rules of long division. Once she demonstrated signs of proficiency in these problems, she moved onto three- and four-digit long division. After she showed progress with those, she was given integrated decimal division. Her homework assignments mirrored what she learned in class, as well as added a few complex problems.

3. Build long-term knowledge. After a month of learning complex multiplication and division, Janine was given an assessment. The assessment required her to write down each step of the process for the multiplication and division problems. She was then given math problems that steadily progressed in difficulty. She improved greatly from the start of our sessions. By starting at the basic level of each process she was able to build the necessary rules to do more complex problems later on. She eventually was teaching herself with ease. By building proper study habits and learning habits she was able to learn and recall information that used to difficult for her.

Recently, there have been several reviews on the efficacy of working memory training (Buschkuehl, Jaeggi, & Jonides, Reference Buschkuehl, Jaeggi and Jonides 2012 ; Jaeggi, Buschkuehl, Jonides, & Shah, Reference Jaeggi, Buschkuehl, Jonides and Shah 2011 ; Morrison & Chein, Reference Morrison and Chein 2011 , for reviews). While there are both positive findings, as well as null results, the key is the working memory program — we cannot apply a blanket statement that all training works (or does not work). Based on research, here are three key things to look for when evaluating the research findings:

1. Control group

A control group offers a comparison to make sure that the training program is not just working because the child is doing something different. Some studies just use a control group who do not do anything. While this is a good start, an ideal control group is a group of people who are doing something different from the training program (such as reading or playing a different computer game). We recently published findings on a working memory training program, Jungle Memory, that included a control group that received additional educational support (Alloway, Reference Alloway 2012a ). The findings indicated that the training group performed significantly better than the control group in standardised tests of IQ and working memory after an 8-week period. Jaeggi et al. ( Reference Jaeggi, Buschkuehl, Jonides and Shah 2011 ) also reported improvements in a non-verbal IQ after a working memory training task ( n -back task) in young adults.

2. Transfer effects

This refers to whether the program improves anything other than getting better at the game itself. Practising one thing will naturally make you better at it. This is known as a practice effect. But can the benefits of a brain training program transfer to real world activities? In other words, can you get better at something other than the training game? In clinical trials, Jungle Memory showed transfer effects — the students showed improvements not just in working memory, but in IQ, and more importantly, in grades as well (Alloway, Reference Alloway 2012a ).

3. Maintenance

How long do the results last? It is important to consider whether the training benefits will last beyond the training period. In a study of almost 100 students, we found that the benefits of Jungle Memory training persisted when students were tested 8 months later (Alloway, Bibile, & Lau, Reference Alloway, Bibile and Lau 2013 ).

In summary, working memory is a foundational skill for learning. It measures our ability to work with information and is linked to learning from kindergarten to the college years. Standardized tests provide accurate and quick ways to assess a student's working memory performance in order to best support their learning. A combination of strategies and training can improve the working memory capacity in our students, thus providing them with an opportunity to reach their potential.

1 Throughout the article the reference to IQ scores is restricted to the Verbal or Performance subscales, rather than the Full Scale IQ score, which does include working memory tests in many of the standardised test batteries (e.g., Wechsler, Woodcock-Johnson, Stanford-Binet).

This article has been cited by the following publications. This list is generated based on data provided by Crossref .

• Google Scholar

View all Google Scholar citations for this article.

Save article to Kindle

To save this article to your Kindle, first ensure [email protected] is added to your Approved Personal Document E-mail List under your Personal Document Settings on the Manage Your Content and Devices page of your Amazon account. Then enter the ‘name’ part of your Kindle email address below. Find out more about saving to your Kindle .

Note you can select to save to either the @free.kindle.com or @kindle.com variations. ‘@free.kindle.com’ emails are free but can only be saved to your device when it is connected to wi-fi. ‘@kindle.com’ emails can be delivered even when you are not connected to wi-fi, but note that service fees apply.

Find out more about the Kindle Personal Document Service.

• Volume 30, Issue 2
• Tracy Packiam Alloway (a1) and Evan Copello (a1)
• DOI: https://doi.org/10.1017/edp.2013.13

Save article to Dropbox

To save this article to your Dropbox account, please select one or more formats and confirm that you agree to abide by our usage policies. If this is the first time you used this feature, you will be asked to authorise Cambridge Core to connect with your Dropbox account. Find out more about saving content to Dropbox .

Save article to Google Drive

To save this article to your Google Drive account, please select one or more formats and confirm that you agree to abide by our usage policies. If this is the first time you used this feature, you will be asked to authorise Cambridge Core to connect with your Google Drive account. Find out more about saving content to Google Drive .

Reply to: Submit a response

- No HTML tags allowed - Web page URLs will display as text only - Lines and paragraphs break automatically - Attachments, images or tables are not permitted

Your details

Your email address will be used in order to notify you when your comment has been reviewed by the moderator and in case the author(s) of the article or the moderator need to contact you directly.

You have entered the maximum number of contributors

Conflicting interests.

Please list any fees and grants from, employment by, consultancy for, shared ownership in or any close relationship with, at any time over the preceding 36 months, any organisation whose interests may be affected by the publication of the response. Please also list any non-financial associations or interests (personal, professional, political, institutional, religious or other) that a reasonable reader would want to know about in relation to the submitted work. This pertains to all the authors of the piece, their spouses or partners.

Working Memory Model (Baddeley and Hitch)

Saul McLeod, PhD

Editor-in-Chief for Simply Psychology

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul McLeod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

Learn about our Editorial Process

Olivia Guy-Evans, MSc

Associate Editor for Simply Psychology

BSc (Hons) Psychology, MSc Psychology of Education

Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.

On This Page:

The Working Memory Model, proposed by Baddeley and Hitch in 1974, describes short-term memory as a system with multiple components.

It comprises the central executive, which controls attention and coordinates the phonological loop (handling auditory information), and the visuospatial sketchpad (processing visual and spatial information).

Later, the episodic buffer was added to integrate information across these systems and link to long-term memory. This model suggests that short-term memory is dynamic and multifaceted.

Take-home Messages

• Working memory is a limited capacity store for retaining information for a brief period while performing mental operations on that information.
• Working memory is a multi-component system that includes the central executive, visuospatial sketchpad, phonological loop, and episodic buffer.
• Working memory is important for reasoning, learning, and comprehension.
• Working memory theories assume that complex reasoning and learning tasks require a mental workspace to hold and manipulate information.

Atkinson’s and Shiffrin’s (1968) multi-store model was extremely successful in terms of the amount of research it generated. However, as a result of this research, it became apparent that there were a number of problems with their ideas concerning the characteristics of short-term memory.

Fig 1 . The Working Memory Model (Baddeley and Hitch, 1974)

Baddeley and Hitch (1974) argue that the picture of short-term memory (STM) provided by the Multi-Store Model is far too simple.

According to the Multi-Store Model , STM holds limited amounts of information for short periods of time with relatively little processing.  It is a unitary system. This means it is a single system (or store) without any subsystems. Whereas working memory is a multi-component system (auditory and visual).

Therefore, whereas short-term memory can only hold information, working memory can both retain and process information.

Working memory is short-term memory . However, instead of all information going into one single store, there are different systems for different types of information.

Central Executive

Visuospatial sketchpad (inner eye), phonological loop.

• Phonological Store (inner ear) processes speech perception and stores spoken words we hear for 1-2 seconds.
• Articulatory control process (inner voice) processes speech production, and rehearses and stores verbal information from the phonological store.

Fig 2 . The Working Memory Model Components (Baddeley and Hitch, 1974)

The labels given to the components (see Fig 2) of the working memory reflect their function and the type of information they process and manipulate.

The phonological loop is assumed to be responsible for the manipulation of speech-based information, whereas the visuospatial sketchpad is assumed to be responsible for manipulating visual images.

The model proposes that every component of working memory has a limited capacity, and also that the components are relatively independent of each other.

The Central Executive

The central executive is the most important component of the model, although little is known about how it functions.  It is responsible for monitoring and coordinating the operation of the slave systems (i.e., visuospatial sketchpad and phonological loop) and relates them to long-term  memory (LTM).

The central executive decides which information is attended to and which parts of the working memory to send that information to be dealt with. For example, two activities sometimes come into conflict, such as driving a car and talking.

Rather than hitting a cyclist who is wobbling all over the road, it is preferable to stop talking and concentrate on driving. The central executive directs attention and gives priority to particular activities.

p> The central executive is the most versatile and important component of the working memory system. However, despite its importance in the working-memory model, we know considerably less about this component than the two subsystems it controls.

Baddeley suggests that the central executive acts more like a system which controls attentional processes rather than as a memory store.  This is unlike the phonological loop and the visuospatial sketchpad, which are specialized storage systems. The central executive enables the working memory system to selectively attend to some stimuli and ignore others.

Baddeley (1986) uses the metaphor of a company boss to describe the way in which the central executive operates.  The company boss makes decisions about which issues deserve attention and which should be ignored.

They also select strategies for dealing with problems, but like any person in the company, the boss can only do a limited number of things at the same time. The boss of a company will collect information from a number of different sources.

If we continue applying this metaphor, then we can see the central executive in working memory integrating (i.e., combining) information from two assistants (the phonological loop and the visuospatial sketchpad) and also drawing on information held in a large database (long-term memory).

The Phonological Loop

The phonological loop is the part of working memory that deals with spoken and written material. It consists of two parts (see Figure 3).

The phonological store (linked to speech perception) acts as an inner ear and holds information in a speech-based form (i.e., spoken words) for 1-2 seconds. Spoken words enter the store directly. Written words must first be converted into an articulatory (spoken) code before they can enter the phonological store.

Fig 3 . The phonological loop

The articulatory control process (linked to speech production) acts like an inner voice rehearsing information from the phonological store. It circulates information round and round like a tape loop. This is how we remember a telephone number we have just heard. As long as we keep repeating it, we can retain the information in working memory.

The articulatory control process also converts written material into an articulatory code and transfers it to the phonological store.

The Visuospatial Sketchpad

The visuospatial sketchpad ( inner eye ) deals with visual and spatial information. Visual information refers to what things look like. It is likely that the visuospatial sketchpad plays an important role in helping us keep track of where we are in relation to other objects as we move through our environment (Baddeley, 1997).

As we move around, our position in relation to objects is constantly changing and it is important that we can update this information.  For example, being aware of where we are in relation to desks, chairs and tables when we are walking around a classroom means that we don”t bump into things too often!

The sketchpad also displays and manipulates visual and spatial information held in long-term memory. For example, the spatial layout of your house is held in LTM. Try answering this question: How many windows are there in the front of your house?

You probably find yourself picturing the front of your house and counting the windows. An image has been retrieved from LTM and pictured on the sketchpad.

Evidence suggests that working memory uses two different systems for dealing with visual and verbal information. A visual processing task and a verbal processing task can be performed at the same time.

It is more difficult to perform two visual tasks at the same time because they interfere with each other and performance is reduced. The same applies to performing two verbal tasks at the same time. This supports the view that the phonological loop and the sketchpad are separate systems within working memory.

The Episodic Buffer

The original model was updated by Baddeley (2000) after the model failed to explain the results of various experiments. An additional component was added called the episodic buffer.

The episodic buffer acts as a “backup” store which communicates with both long-term memory and the components of working memory.

Fig 3 . Updated Model to include the Episodic Buffer

Critical Evaluation

Researchers today generally agree that short-term memory is made up of a number of components or subsystems. The working memory model has replaced the idea of a unitary (one part) STM as suggested by the multistore model.

The working memory model explains a lot more than the multistore model. It makes sense of a range of tasks – verbal reasoning, comprehension, reading, problem-solving and visual and spatial processing. The model is supported by considerable experimental evidence.

The working memory applies to real-life tasks:

• reading (phonological loop)
• problem-solving (central executive)
• navigation (visual and spatial processing)

The KF Case Study supports the Working Memory Model. KF suffered brain damage from a motorcycle accident that damaged his short-term memory.

KF’s impairment was mainly for verbal information – his memory for visual information was largely unaffected. This shows that there are separate STM components for visual information (VSS) and verbal information (phonological loop).

The working memory model does not over-emphasize the importance of rehearsal for STM retention, in contrast to the multi-store model.

Empirical Evidence for Working Memory

What evidence is there that working memory exists, that it comprises several parts, that perform different tasks? Working memory is supported by dual-task studies (Baddeley and Hitch, 1976).

The working memory model makes the following two predictions:

1 . If two tasks make use of the same component (of working memory), they cannot be performed successfully together. 2 . If two tasks make use of different components, it should be possible to perform them as well as together as separately.

Key Study: Baddeley and Hitch (1976)

Aim : To investigate if participants can use different parts of working memory at the same time.

Method : Conducted an experiment in which participants were asked to perform two tasks at the same time (dual task technique) – a digit span task which required them to repeat a list of numbers, and a verbal reasoning task which required them to answer true or false to various questions (e.g., B is followed by A?).

Results : As the number of digits increased in the digit span tasks, participants took longer to answer the reasoning questions, but not much longer – only fractions of a second.  And, they didn”t make any more errors in the verbal reasoning tasks as the number of digits increased.

Conclusion : The verbal reasoning task made use of the central executive and the digit span task made use of the phonological loop.

Brain Imaging Studies

Several neuroimaging studies have attempted to identify distinct neural correlates for the phonological loop and visuospatial sketchpad posited by the multi-component model.

For example, some studies have found that tasks tapping phonological storage tend to activate more left-hemisphere perisylvian language areas, whereas visuospatial tasks activate more right posterior regions like the parietal cortex (Smith & Jonides, 1997).

However, the overall pattern of results remains complex and controversial. Meta-analyses often fail to show consistent localization of verbal and visuospatial working memory (Baddeley, 2012).

There is significant overlap in activation, which may reflect binding processes through the episodic buffer, as well as common executive demands.

Differences in paradigms and limitations of neuroimaging methodology further complicate mapping the components of working memory onto distinct brain regions or circuits (Henson, 2001).

While neuroscience offers insight into working memory, Baddeley (2012) argues that clear anatomical localization is unlikely given the distributed and interactive nature of working memory. Specifically, he suggests that each component likely comprises a complex neural circuit rather than a circumscribed brain area.

Additionally, working memory processes are closely interrelated with other systems for attention, perception and long-term memory . Thus, neuroimaging provides clues but has not yet offered definitive evidence to validate the separable storage components posited in the multi-component framework.

Further research using techniques with higher spatial and temporal resolution may help better delineate the neural basis of verbal and visuo-spatial working memory.

Lieberman (1980) criticizes the working memory model as the visuospatial sketchpad (VSS) implies that all spatial information was first visual (they are linked).

However, Lieberman points out that blind people have excellent spatial awareness, although they have never had any visual information. Lieberman argues that the VSS should be separated into two different components: one for visual information and one for spatial.

There is little direct evidence for how the central executive works and what it does. The capacity of the central executive has never been measured.

Working memory only involves STM, so it is not a comprehensive model of memory (as it does not include SM or LTM).

The working memory model does not explain changes in processing ability that occur as the result of practice or time.

State-based models of WM

Early models of working memory proposed specialized storage systems, such as the phonological loop and visuospatial sketchpad, in Baddeley and Hitch’s (1974) influential multi-component model.

However, newer “state-based” models suggest working memory arises from temporarily activating representations that already exist in your brain’s long-term memory or perceptual systems.

For example, you activate your memory of number concepts to remember a phone number. Or, to remember where your keys are, you activate your mental map of the room.

According to state-based models, you hold information in mind by directing your attention to these internal representations. This gives them a temporary “boost” of activity.

More recent state-based models argue against dedicated buffers, and propose that working memory relies on temporarily activating long-term memory representations through attention (Cowan, 1995; Oberauer, 2002) or recruiting perceptual and motor systems (Postle, 2006; D’Esposito, 2007).

Evidence from multivariate pattern analysis (MVPA) of fMRI data largely supports state-based models, rather than dedicated storage buffers.

For example, Lewis-Peacock and Postle (2008) showed MVPA classifiers could decode information being held in working memory from patterns of activity associated with long-term memory for that content.

Other studies have shown stimulus-specific patterns of activity in sensory cortices support the retention of perceptual information (Harrison & Tong, 2009; Serences et al., 2009).

Atkinson, R. C., & Shiffrin, R. M. (1968). Chapter: Human memory: A proposed system and its control processes. In Spence, K. W., & Spence, J. T. The psychology of learning and motivation (Volume 2). New York: Academic Press. pp. 89–195.

Baddeley, A. D. (1986). Working memory . Oxford: Oxford University Press.

Baddeley, A. (1996). Exploring the central executive.  The Quarterly Journal of Experimental Psychology Section A ,  49 (1), 5-28.

Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences , 4, (11): 417-423.

Baddeley, A. (2010). Working memory.  Current biology ,  20 (4), R136-R140.

Baddeley, A. (2012). Working memory: Theories, models, and controversies.  Annual review of psychology ,  63 , 1-29.

Baddeley, A. D., & Hitch, G. (1974). Working memory. In G.H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 8, pp. 47–89). New York: Academic Press.

Baddeley, A. D., & Lieberman, K. (1980). Spatial working memory. ln R. Nickerson. Attention and Performance, VIII . Hillsdale, N): Erlbaum.

Borella, E., Carretti, B., Sciore, R., Capotosto, E., Taconnat, L., Cornoldi, C., & De Beni, R. (2017). Training working memory in older adults: Is there an advantage of using strategies?.  Psychology and Aging ,  32 (2), 178.

Chai, W. J., Abd Hamid, A. I., & Abdullah, J. M. (2018). Working memory from the psychological and neurosciences perspectives: a review.  Frontiers in Psychology ,  9 , 401.

Cowan, N. (1995). Attention and memory: An integrated framework . Oxford Psychology Series, No. 26. New York: Oxford University Press.

Cowan, N. (2005). Working memory capacity.  Exp. Psychology,  54, 245–246.

Cowan, N. (2008). What are the differences between long-term, short-term, and working memory?.  Progress in brain research ,  169 , 323-338.

Curtis, C.E., & D’Esposito, M. (2003). Persistent activity in the prefrontal cortex during working memory. Trends in Cognitive Sciences, 7 (9), 415-423.

D’Esposito, M. (2007). From cognitive to neural models of working memory. Philosophical Transactions of the Royal Society B: Biological Sciences, 362 (1481), 761-772.

D’Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory.  Annual review of psychology ,  66 , 115-142.

Fell, J., & Axmacher, N. (2011). The role of phase synchronization in memory processes. Nature Reviews Neuroscience, 12 (2), 105-118.

Harrison, S. A., & Tong, F. (2009). Decoding reveals the contents of visual working memory in early visual areas. Nature, 458 (7238), 632-635.

Henson, R. (2001). Neural working memory. In J. Andrade (Ed.), Working memory in perspective (pp. 151-174). Psychology Press.

Lewis-Peacock, J. A., & Postle, B. R. (2008). Temporary activation of long-term memory supports working memory. Journal of Neuroscience, 28 (35), 8765-8771.

Lieberman, M. D. (2000). Introversion and working memory: Central executive differences.  Personality and Individual Differences ,  28 (3), 479-486.

Osaka, M., Osaka, N., Kondo, H., Morishita, M., Fukuyama, H., Aso, T., & Shibasaki, H. (2003). The neural basis of individual differences in working memory capacity: an fMRI study.  NeuroImage ,  18 (3), 789-797.

Serences, J.T., Ester, E.F., Vogel, E.K., & Awh, E. (2009). Stimulus-specific delay activity in human primary visual cortex. Psychological Science, 20( 2), 207-214.

Smith, E.E., & Jonides, J. (1997). Working memory: A view from neuroimaging. Cognitive Psychology, 33 (1), 5-42.

Publication in Journal of Political Economy

13 September 2024 12:00

A paper by Henning Hermes, Eva M. Berger, Daniel Schunk, Kirsten Winkel titled "The Impact of Working Memory Training on Children’s Cognitive and Noncognitive Skills" is published in Journal of Political Economy.

Working memory (WM) capacity is a key component of a wide range of cognitive and noncognitive skills, such as fluid IQ, math, reading, or inhibitory control - but can WM traning improve these skills? Here, we examine the casual impact of WM training embedded in regular school teaching based on a randomized educational intervention with 6-7 y old children.  We find substantial fains in WM capacity, and document positive spillover effects on geometry, fluid IQ, and inhibitory control. Three years later, treated children are 16 percentage points more likely to enter an advanced secondary school track.

Read the paper here

• Share on Facebook
• Share on Twitter
• Share on LinkedIn
• Share by e-mail

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals

Long-term memory articles from across Nature Portfolio

Long-term memory is information encoded in the brain on the time-scale of years. It consists of explicit (declarative) memories that are consciously reportable and depend heavily on the medial temporal lobe and hippocampus and implicit (procedural) memories that are unconscious and depend on the basal ganglia and cerebellum.

Latest Research and Reviews

Co-administration of midazolam and psilocybin: differential effects on subjective quality versus memory of the psychedelic experience

• Christopher R. Nicholas
• Matthew I. Banks
• Charles L. Raison

Embodiment in episodic memory through premotor-hippocampal coupling

This study describes a premotor-hippocampal coupling in episodic memory, revealing how the bodily sensory context and in particular the sense of agency of the observer at encoding is neurally reinstated during the retrieval of past episodes.

• Nathalie Heidi Meyer
• Baptiste Gauthier
• Olaf Blanke

Concept and location neurons in the human brain provide the ‘what’ and ‘where’ in memory formation

Whether specialized neuronal firing in the human MTL predicts successful memory encoding remains unknown. Here, the authors find this to be the case for two distinct populations of single neurons responding to items and locations, respectively.

• Sina Mackay
• Thomas P. Reber
• Florian Mormann

Active intentional and unintentional forgetting in the laboratory and everyday life

Forgetting can be an intentional act or an unintentional side effect of other memory processes. In this Review, Fawcett et al. describe intentional and unintentional forgetting in laboratory and real-world settings, highlight alignment across these traditionally siloed research areas, and explore the role of forgetting in certain clinical conditions.

• Jonathan M. Fawcett
• Tracy L. Taylor
• Ashleigh M. Maxcey

Grasp and remember: the impact of human and robotic actions on object preference and memory

• Alex Kafkas
• Megan Rowland
• Luca F. Ticini

Arc and BDNF mediated effects of hippocampal astrocytic glutamate uptake blockade on spatial memory stages

Modulation of glutamate uptake in the rat hippocampus via astroglial GLT-1 improves or impairs spatial objects memory depending on the intensity of learning session as well as the memory phase evaluated, being Arc and BDNF mediators of these effects.

• Juan Gabriel Riboldi
• Julieta Correa
• Haydee Viola

News and Comment

Dopaminergic and noradrenergic nuclei relate to different memory types in older adults

By using multimodal MRI in aging humans, we found that noradrenergic brain regions are associated with episodic memory impairment, whereas dopaminergic areas are implicated in working memory impairment. Unravelling the role of changes in these neurotransmitter systems in age-related memory loss contributes to our understanding of the development of neurodegenerative diseases.

Fear memory engram is the mind-killer

Stegemann et al. have unraveled a long-lasting role for fear engrams in future pain perception. Prefrontal pain and fear representations become entangled following a fearful event. As subsequent painful experiences also reactivate the fear network, silencing the fear engram alone can alleviate both tonic pain and hypersensitivity in chronic pain.

• Clément Pouget
• Gisella Vetere

Retroactive interference arises from semantic similarity

• Teresa Schubert

A frequency location to remember

Neuromodulation with specific frequencies at specific brain locations selectively enhances either working memory or long-term memory in older adult humans.

• Jake Rogers

Openness about animal research increases public support

Science engagement can be a daunting prospect. This is especially true for scientists whose work involves animal models, and particularly nonhuman primates. Here, we show that openly explaining our rationale for our neuroscience work involving nonhuman primates — and the legal and ethical regulations that govern animal experimentation — increased public support and understanding, which is crucial for this essential research to continue.

• Juan Carlos Mendez
• Brook A. L. Perry
• Anna S. Mitchell

A distinct signaling pathway in parvalbumin-positive interneurons controls flexible memory updating

• Denise J. Cai
• Tristan Shuman

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

• About the DPRU
• Staff Profiles
• Prof. Haroon Bhorat - profile
• Social Responsiblity
• Publications Overview
• Working Papers
• Policy Briefs
• Fact Sheets
• Other Publications
• Research Overview
• Brookings Africa Growth Initiative
• Minimum Wages
• COVID-19: Labour Markets & Social Assistance
• INSETA Research Chair
• merSETA: Economic Complexity
• Towards Resilient Futures: CoP phase 2
• EPP Phase III: Research, Projects & Outputs
• ELLA: Experiences and Learning from Latin America
• First IZA / WB / NJD Conference on Jobs & Development
• China’s Impact on African Employment: What Do We Know and Where are the Gaps?
• Africa's Lions
• Building economic complexity in Africa
• Towards Resilient Futures: CoP phase 1
• Multimedia Overview
• Media Clippings
• Current News
• Archived News

In Memory of Minister Pravin Gordhan

On behalf of Prof. Haroon Bhorat, Director of the DPRU, we share heartfelt sympathies following the passing of former Minister Pravin Gordhan.

Prof. Bhorat served as an economic advisor to Gordhan during his tenure as Minister of Finance. “He was one of the greatest South Africans in the modern era… this is a very sad moment for the country.”

It is with a deep sense of loss that we extend our sincere condolences to his family, friends and colleagues.

"The acts of the greedy, the corrupt, the bully, the counter-revolutionary set back our progress as a democracy and stop us from becoming a caring nation. While they sit back to enjoy their spoils, the damage they cause is borne by our communities – by workers, by small businesses, by the unemployed and youth. Now is the time for all of us to join the ranks of those who want to build a better future and better institutions and not just point fingers among us. Strong, organised communities are fundamental to the security of our country’s infrastructure."

Pravin Gordhan,  Budget Vote Speech, 2022