Modeling working memory: a computational implementation of the Time-Based Resource-Sharing theory

Working memory is a core concept in cognition, predicting about 50% of the variance in IQ and reasoning tasks. A popular test of working memory is the complex span task, in which encoding of memoranda alternates with processing of distractors. A recent model of complex span performance, the Time-Based-Resource-Sharing (TBRS) model of Barrouillet and colleagues, has seemingly accounted for several crucial findings, in particular the intricate trade-off between deterioration and restoration of memory in the complex span task. According to the TBRS, memory traces decay during processing of the distractors, and they are restored by attentional refreshing during brief pauses in between processing steps. However, to date, the theory has been formulated only at a verbal level, which renders it difficult to test and to be certain of its intuited predictions. We present a computational instantiation of the TBRS and show that it can handle most of the findings on which the verbal model was based. We also show that there are potential challenges to the model that await future resolution. This instantiated model, TBRS*, is the first comprehensive computational model of performance in the complex span paradigm. The Matlab model code is available as a supplementary material of this article.     

The effects of prefrontal lesions on working memory performance and theory

The effects of experimental lesions of the monkey prefrontal cortex have played a predominant role in current conceptualizations of the functional organization of the lateral prefrontal cortex, especially with regard to working memory. The loss or sparing of certain performance abilities has been shown to be attributable to differences in the specific requirements of behavioral testing (e.g., spatial vs. nonspatial memoranda) along with differences in the specific locations of applied ablations (e.g., dorsal vs. ventral prefrontal cortex). Such findings, which have accumulated now for over a century, have led to widespread acceptance that the dorsolateral and ventrolateral aspects of the prefrontal cortex may perform different, specialized roles in higher order cognition. Nonetheless, it remains unclear and controversial how the lateral prefrontal cortex is functionally organized. Two main views propose different types of functional specialization of the dorsal and ventral prefrontal cortex. The first contends that the lateral prefrontal cortex is segregated according to the processing of spatial and nonspatial domains of information. The second contends that domain specialization is not the key to the organization of the prefrontal cortex, but that instead, the dorsal and ventral prefrontal cortices perform qualitatively different operations. This report critically reviews all relevant monkey lesion studies that have served as the foundation for current theories regarding the functional organization of the prefrontal cortex. Our goals are to evaluate how well the existing lesion data support each theory and to enumerate caveats that must be considered when interpreting the relevant literature.     

A capacity theory of comprehension: individual differences in working memory.

A theory of the way working memory capacity constrains comprehension is proposed. The theory proposes that both processing and storage are mediated by activation and that the total amount of activation available in working memory varies among individuals. Individual differences in working memory capacity for language can account for qualitative and quantitative differences among college-age adults in several aspects of language comprehension. One aspect is syntactic modularity: The larger capacity of some individuals permits interaction among syntactic and pragmatic information, so that their syntactic processes are not informationally encapsulated. Another aspect is syntactic ambiguity: The larger capacity of some individuals permits them to maintain multiple interpretations. The theory is instantiated as a production system model in which the amount of activation available to the model affects how it adapts to the transient computational and storage demands that occur in comprehension.     

Implicit working memory

Working Memory (WM) plays a crucial role in many high-level cognitive processes (e.g., reasoning, decision making, goal pursuit and cognitive control). The prevalent view holds that active components of WM are predominantly intentional and conscious. This conception is oftentimes expressed explicitly, but it is best reflected in the nature of major WM tasks: All of them are blatantly explicit. We developed two new WM paradigms that allow for an examination of the role of conscious awareness in WM. Results from five studies show that WM can operate unintentionally and outside of conscious awareness, thus suggesting that the current view should be expanded to include implicit WM.     

Distinguishing short-term memory from working memory

The aim of the present research was to determine whether short-term memory and working memory could be distinguished. In two studies, 7- to 13-year-olds (N = 155, N = 132) were administered tasks thought to assess short-term memory as well as tasks thought to assess working memory. Both exploratory and confirmatory factor analyses distinguished short-term memory tasks from working memory tasks. In addition, performance on working memory tasks was related to word decoding skill but performance on short-term memory tasks was not. Finally, performance on both short-term memory and working memory tasks were associated with age-related increases in processing speed. Results are discussed in relation to models of short-term and working memory.     

A source activation theory of working memory: cross-task prediction of performance in ACT-R

Over the decades, computational models of human cognition have advanced from programs that produce output similar to that of human problem solvers to systems that mimic both the products and processes of human performance. In this paper, we describe a model that achieves the next step in this progression: predicting individual participants’ performance across multiple tasks after estimating a single individual difference parameter. We demonstrate this capability in the context of a model of working memory, where the individual difference parameter for each participant represents working memory capacity. Specifically, our model is able to make zero-parameter predictions of individual participants’ performance on a second task after separately fitting performance on a preliminary task. We argue that this level of predictive ability offers an important test of the theory underlying our model.     

Visual-spatial working memory, attention, and scene representation: A neuro-cognitive theory

This paper addresses the issue of how visual-spatial working memory, attention, and scene representation are related. The first section introduces a modified two-stage conception of visual-spatial processing. “Stage one” refers to low-level visual-spatial processing and computes in parallel for the currently available retinal information “object candidates,” here called “visual-spatial units.” An attentional process called “unit selection” allows access to stage two for one of these units at a time. Stage two contains high-level visual-spatial information that can be used for goal-directions (e.g., verbal report, grasping). It consists of three parallel processing streams. First, the currently selected unit is recognized; second, a spatial-motor program for the selected unit is computed; and third, an “object file” is set up for the selected unit. An object file contains temporary episodic representations of detailed high-level visual-spatial attributes of an “object” plus an “index.” An index acts as a pointer and is bound via temporary connections to the attributes of the file. Section two of this paper specifies one part of stage two in more detail, namely visual-spatial working memory (VSWM). It can contain up to four object files. A first central claim is that during sensory-based processing for working memory (“access”), one object file is always “on-line,” and up to three other object files are “off-line”. A second central claim is that the process of setting up an object file depends on the number and the activation level of already stored files. Based on the concept of activation-based competition between object files, it is postulated that the more files that are stored and the higher their activation is, the longer it takes for a newly set up object file to reach a sufficient level of activation. Activation-based competition is also used to explain “short-term forgetting” by “interference.” A third central claim about VSWM is that a “refreshment' process exists that increases the activation level of an index of an object file in order to prevent forgetting or in order to bring the file back to the state of controlling the current action. Finally, section three gives a selective look at a number of experimental data such as the attentional blink, backward masking, dwell time effects, transsaccadic memory, and change blindness. New explanations are offered and new predictions made. Received: 16 January 1998 / Accepted: 25 July 1998     

Effect of working memory training on working memory,arithmetic and following instructions

Mathematical ability is dependent on specific mathematical training but also associated with a range of cognitive factors, including working memory (WM) capacity. Previous studies have shown that WM training leads to improvement in non-trained WM tasks, but the results regarding transfer to mathematics are inconclusive. In the present study, 176 children with WM deficits, aged 7–15 years performed 5 weeks of WM training. During the training period, they were assessed five times with a test of complex WM (the Odd One Out), a test of remembering and following instructions and a test of arithmetic. The improvements were compared to the performance of a control group of 304 typically developing children aged 7–15 years who performed the same transfer tasks at the same time intervals, but without training. The training group improved significantly more than the control group on all three transfer tests (all p < 0.0001), after correction for baseline performance, age and sex. The effect size for mathematics was small and the effect sizes for the WM tasks were moderate to large. The transfer increased linearly with the amount of training time and correlated with the amount of improvement on the trained tasks. These results confirm previous findings of training-induced improvements in non-trained WM tasks including the ability to follow instructions, but extend previous findings by showing improvements also for arithmetic. This is encouraging regarding the potential role of cognitive training for education, but it is desirable to find paradigms that would enhance the effect of the training on mathematics. One of the future challenges for studying training effects is combining large sample sizes with high quality and compliance, to detect relevant but smaller effects of cognitive training.     

Dementia and working memory

This study explored the hypothesis that patients suffering from dementia of the Alzheimer type (DAT) are particularly impaired in the functioning of the Central Executive component of working memory, and that this will be reflected in the capacity of patients to perform simultaneously two concurrent tasks. DAT patients, age-matched controls and young controls were required to combine performance on a tracking task with each of three concurrent tasks, articulatory suppression, simple reaction time to a tone and auditory digit span. The difficulty of the tracking task and length of digit sequence were both adjusted so as to equate performance across the three groups when the tasks were performed alone. When digit span or concurrent RT were combined with tracking, the deterioration in performance shown by the DAT patients was particularly marked.     

The development of working memory: further note on the comparability of two models of working memory

This paper is a set of reflections on Kemps, De Rammelaere, and Desmet's article (2000, this issue), in which the two models by Baddeley and Pascual-Leone are compared. First, some of the similarities and differences between the two models which we identified in a 1994 paper (de Ribaupierre & Bailleux, 1994) are briefly summarized and reexamined in the light of more recent work. Second, we debate the issue of whether each model makes a specific contribution to the explanation of some of Kemps et al.'s results, that is, of whether they can be considered to be complementary. Third, we argue for the necessity of theoretical task analyses, in view of the divergent results obtained in the two tasks used (the Corsi and the Peanut tasks), notably different developmental profiles, and an overall higher level of performance in the Corsi task. Finally, we briefly summarize a very similar study in which we also used Mr. Peanut with concurrent tasks in children and in young adults and in which we obtained rather different results. By comparing the experimental procedures used in the two studies, we contribute some exploratory hypotheses, while raising issues that can easily be generalized to other visuo-spatial working memory tasks.     

On the nonassociative nature of working memory

Rats were trained in an eight-arm radial maze in two experiments. A “fixed set” of four arms, randomly selected and sequenced, was assigned to each rat. The fixed set was always presented in the same order. Each training trial began with rewarded forced choices of the four arms in the fixed set. The trial concluded with a free-choice test among all eight arms after a delay of about 40 min; choices of the four arms not in the fixed set were rewarded. The degree to which the identities and order of the arms in the fixed set had been learned was assessed in a reversal test. The four arms that were not in the fixed set were presented at the beginning of the reversal test trial; after the usual delay a free-choice test was given, but choices of arms in the fixed set were rewarded. In both experiments, the rats showed no negative transfer in the reversal test. Moreover, sequences of postdelay choices during training varied across days, and the sequence of choices during the reversal test did not match the sequence of arms in the fixed set. The rats therefore learned little if anything about the identities and the order of arms in the fixed set in spite of the consistencies between trials and the maintenance of the fixed set in memory for 40 min within trials. The finding is congruent with the view of working memory as nonassociative.     

Roles of working memory capacity and long-term working memory skill in complex task performance

In the present research, we examined the relative roles of domain-general and domain-specific individual difference characteristics in complex cognitive task performance. Specifically, we examined the impact both of working memory (WM) capacity and of acquired skills used to encode presented information in an accessible form in long-term working memory (LTWM) on performance in a complex aviation task environment. Measures of WM capacity and LTWM skill served as performance predictors. A criterion measure of task performance was related to the predictor measures. The results indicated that an increase in LTWM skill decreases the role of WM capacity as the determinant of complex task performance, although both measures are important performance predictors. We discuss how the two distinct WM constructs coexist and interact to support complex task performance.     

Conceptual similarity effects on working memory in sentence contexts: Testing a theory of anaphora

The degree of semantic similarity between an anaphoric noun phrase (e.g., the bird) and its antecedent (e.g., a robin) is known to affect the anaphor resolution process, but the mechanisms that underlie this effect are not known. One proposal (Almor, 1999) is that semantic similarity triggers interference effects in working memory and makes two crucial assumptions: First, semantic similarity impairs working memory just as phonological similarity does (e.g., Baddeley, 1992), and, second, this impairment interferes with processes of sentence comprehension. We tested these assumptions in two experiments that compared recall accuracy between phonologically similar, semantically similar, and control words in sentence contexts. Our results do not provide support for Almor's claims: Phonological overlap decreased recall accuracy in sentence contexts, but semantic similarity did not. These results shed doubt on the idea that semantic interference in working memory is an underlying mechanism in anaphor resolution.     

Investigating the other-race effect in working memory

People have difficulties in remembering other-race faces; this so-called other-race effect (ORE) has been frequently observed in long-term recognition memory (LTM). Several theories argue that the ORE in LTM is caused by differences in earlier processing stages, such as encoding of ingroup and outgroup faces. We test this hypothesis by exploring whether the ORE can already be observed in visual working memory (VWM)—an intermediate system located between encoding processes and LTM storage. In four independent experiments, we observed decreased performance for outgroup faces compared to ingroup faces using three different VWM tasks: an adaptive N-back task, a self-ordered pointing task, and a change detection task. Also, we found that the number of items stored in VWM is smaller for outgroup faces than for ingroup faces. Further, we explored whether performance differences in the change detection task are related to the classic ORE in recognition memory. Our results provide further evidence that the ORE originates during earlier stages of cognitive processing. We discuss that (how) future ORE research may benefit from considering theories and evidence from the VWM literature.     

How rats perform spatial working memory tasks: limitations in the use of egocentric and idiothetic working memory

Rats of the Dark Agouti strain were trained on delayed alternation under conditions that should encourage egocentric working memory. In two experiments a T-maze was set within a cross-maze so that different arms could be used for the sample and test runs. The maze had high opaque side-walls, and testing was conducted in low light levels so that distal visual cues might be eliminated. By rotating the maze 90° between the sample and choice run and by using two identical mazes set side by side it was possible to nullify other spatial strategies. Experiments 1 and 2 showed that rats preferentially used place information, intramaze cues, and direction cues, even though only egocentric or idiothetic (nonmatch-to-turn) working memory could successfully solve every trial. Rats were able to maintain an accurate sense of location within the maze even though distal cues were not visible and the animal was moved between the sample and choice runs. Experiment 2 confirmed that another rat strain (Long-Evans) shows the same learning profiles. Both experiments indicate that rats are very poor at using either egocentric or idiothetic information to alternate, and that retention delays as short as 10 s can eliminate the use of these forms of memory.     

How rats perform spatial working memory tasks: Limitations in the use of egocentric and idiothetic working memory

Rats of the Dark Agouti strain were trained on delayed alternation under conditions that should encourage egocentric working memory. In two experiments a T-maze was set within a cross-maze so that different arms could be used for the sample and test runs. The maze had high opaque side-walls, and testing was conducted in low light levels so that distal visual cues might be eliminated. By rotating the maze 90° between the sample and choice run and by using two identical mazes set side by side it was possible to nullify other spatial strategies. Experiments 1 and 2 showed that rats preferentially used place information, intramaze cues, and direction cues, even though only egocentric or idiothetic (nonmatch-to-turn) working memory could successfully solve every trial. Rats were able to maintain an accurate sense of location within the maze even though distal cues were not visible and the animal was moved between the sample and choice runs. Experiment 2 confirmed that another rat strain (Long-Evans) shows the same learning profiles. Both experiments indicate that rats are very poor at using either egocentric or idiothetic information to alternate, and that retention delays as short as 10 s can eliminate the use of these forms of memory.     

Sentence production and working memory

Previous research has shown that subjects can make up sentences faster from related noun pairs than from unrelated pairs. Two experiments tested the proposal that for the related pairs a subject simply has to access stored information but must make up an appropriate relation for unrelated pairs. In both studies the subjects were asked on some trials to remember a digit preload while they made up their sentences. Contrary to an initial prediction, the latencies were faster rather than slower with a preload. Furthermore, a surprise recall test at the end of Experiment 2 showed that there was no difference in the recall of related and unrelated pairs. The results were interpreted as evidence that semantic work was necessary for both types of noun pairs, and that the subjects did less semantic work with a digit preload. This interpretation was further supported by the finding that with the preload the subjects produced a narrower range of semantic relations between the noun pairs.     

The primate working memory networks

Working memory has long been associated with the prefrontal cortex, since damage to this brain area can critically impair the ability to maintain and update mnemonic information. Anatomical and physiological evidence suggests, however, that the prefrontal cortex is part of a broader network of interconnected brain areas involved in working memory. These include the parietal and temporal association areas of the cerebral cortex, cingulate and limbic areas, and subcortical structures such as the mediodorsal thalamus and the basal ganglia. Neurophysiological studies in primates confirm the involvement of areas beyond the frontal lobe and illustrate that working memory involves parallel, distributed neuronal networks. In this article, we review the current understanding of the anatomical organization of networks mediating working memory and the neural correlates of memory manifested in each of their nodes. The neural mechanisms of memory maintenance and the integrative role of the prefrontal cortex are also discussed.