An Introduction to Applied Cognitive Psychology - David Groome, Anthony Esgate, Michael W. Eysenck (2016)
Chapter 5. Working memory and performance limitations
WORKING MEMORY AND EVERYDAY COGNITION
The performance of many everyday cognitive tasks requires the short-term retention and simultaneous manipulation of material. Psychologists use the term ‘working memory’ to refer to the system responsible for the temporary storage and concurrent processing of information. Indeed, working memory has been described as ‘memory in the service of cognition’ (Salthouse, 2015). Early models of short-term memory generally ignored its function in everyday cognition where the processing of temporarily stored information is often essential to task performance (Mathews et al., 2000). Working memory appears to be particularly vulnerable to the effects of ageing, with significant consequences for real-world cognition in older people. Recent research has examined interventions aimed at minimising the effects of cognitive ageing through working memory training.
Given its apparent centrality in cognition, it is not surprising that working memory plays an important role in a number of cognitive functions, including comprehension, learning, reasoning, problem solving and reading (Shah and Miyake, 1999). Indeed, the working memory frame work has been usefully applied to a range of real-world tasks. The activities discussed in this chapter comprise air traffic control, mental calculation, learning programming languages, cognitive tasks in older people and human–computer interaction. These activities were selected because they provide a clear demonstration of the utility of the working memory concept in understanding performance limitations in real-world tasks.
MODELS OF WORKING MEMORY
There are a number of differing models of working memory. In North America, the emphasis has largely been on working memory as a general attentional processor applied to the temporary activation of unitary long-term memory (e.g. Cowan, 2005). In Europe, short-term multi-component models comprising modality-specific components have been favoured. Arguably the most influential model of working memory has been developed by Alan Baddeley and his collaborators (e.g. Baddeley and Hitch, 1974; Baddeley, 1986; Baddeley and Logie, 1999; Baddeley et al., 2010; Baddeley, 2012). In this model the working memory system is seen as having at least four components: a supervisory attention-controlling ‘central executive’, a speech-based ‘phonological loop’, a ‘visuo spatial sketchpad’ and an ‘episodic buffer’ (see Figure 5.1).
Each of the components of working memory has limited capacity which is reflected in the limitations of human performance exhibited in various working memory tasks. The central executive ‘man ages’ working memory by executing a number of control processes. Examples of executive control processes are: maintaining and updating task goals, monitoring and correcting errors, scheduling responses, initiating rehearsal, inhibiting irrelevant information, retrieving information from long-term memory, switching retrieval plans, and coordinating activity in concurrent tasks. The central executive also coordinates the activity of the phonological loop and the visuospatial sketchpad. The phono logical loop is a speech-based processor consisting of a passive storage device, the ‘phonological store’, coupled to an active subvocal rehearsal mechanism known as the ‘articulatory loop’ (Baddeley, 1997). It is responsible for the short-term retention of material coded in a phonological format. The visuospatial sketchpad (VSSP) retains information coded in a visuospatial form. Recently a fourth component has been added to Baddeley’s model: the ‘episodic buffer’ (see Figure 5.1). The episodic buffer provides a means of linking working memory to long-term memory and perception in addition to providing buffer storage for the components of working memory to communicate with each other (see Baddeley, 2012). These are essential functions in many everyday cognitive tasks. Indeed, Baddeley regards working memory as an interactive system that links incoming perceptual information to long-term memory, thereby providing an interface between cognition and action (Baddeley, 2012; see Figure 5.2).
Figure 5.1 A model of working memory based on Baddeley (2012).
As we have seen, Baddeley’s model assumes separate but interacting short-term and long-term memory systems. As such, it can be distinguished from models that posit that working memory is not a separate memory system but rather the temporary activation of long-term memory. For example, in his ‘embedded processes theory’, Cowan argues that working memory is essentially the application of focused attention to areas of long-term memory (e.g. Cowan, 2005). For Cowan, the capacity of working memory is determined by the limited capacity of attention, and this is limited to a maximum of four discrete chunks. Baddeley agrees with Cowan in as much as he sees the capacity of the episodic buffer as being limited to approximately four chunks of information, but each chunk may contain more than one item (Baddeley, 2012). Indeed, despite differences in their use of terminology, Baddeley regards Cowan’s model as being largely consistent with his own; embedded processes theory is concerned with the relationship between the central executive and the episodic buffer, although Cowan does not use these terms. In addition, Baddeley regards long-term memory as contributing to the operation of working memory components such as the phono logical loop, but points out that evidence of neuropsychological dissociations highlights the theoretical value of retaining a separate short-term working memory system in his model.
Figure 5.2 A model of the relationship between working memory, long-term memory, perception and action based on Baddeley (2012).
5.2 WORKING MEMORY AND AGEING
An important real-world phenomenon to which the working memory framework has been usefully applied is cognitive ageing. Indeed, working memory has been identified as the key source of age-related deficits in a range of cognitive tasks, including long-term memory, language, problem solving and decision making, and the majority of theories of cognitive ageing implicate working memory (Glisky, 2007). Work in this area may have applied value because one of the principal objectives of this research is to use the knowledge gained to develop effective interventions to minimise or reverse the effects of ageing on mental functioning (Braver and West, 2008).
Tasks on which older adults demonstrate impairments relative to younger adults include those that require the active manipulation of the contents of short-term memory; consistent with this, many studies indicate that it is working memory rather than passive short-term memory per se that is most vulnerable to the effects of both normal and abnormal ageing (Baddeley, 1986; Glisky, 2007). Specifically, the effects of ageing are most pronounced in cognitive tasks that require the dynamic, flexible control of attention; these are functions associated with the central executive of working memory. Neuroimaging studies reveal increased activity in the dorsolateral area of the prefrontal cortex (PFC) of persons engaged in such tasks, and it therefore appears that this part of the PFC provides the neural basis of executive control in working memory. In contrast, the information maintenance element of working memory tasks activates the ventrolateral PFC. Studies have demonstrated that when tasks require executive control, older participants show higher levels of dorsolateral PFC activation than younger adults, and this appears to reflect diminished neural efficiency in this area, requiring greater activation levels to achieve executive control (Hedden, 2007).
The executive control of attention is required in tasks that require the updating of information in working memory, task switching or the inhibition of irrelevant information, for example. With respect to the latter, diminution of inhibitory control has been identified by Lynn Hasher and her colleagues as a key factor in the poorer performance of older participants in cognitive tasks (e.g. Hasher et al., 1999). Hasher et al.’s conclusions are based in part on the absence of negative priming effects seen in several studies of elderly participants. Negative priming refers to the disruption (typically slowing) of a response to an item if that item has been previously ignored (Tipper, 2001). Thus negative priming is assumed to reflect an attentional inhibition mechanism. More recent work has suggested possible alternative explanations of the negative priming effect, but converging evidence for diminished inhibitory control in older persons has also come from a variety of other empirical methods. For example, Lustig et al. (2001) compared old and young adults on working-span tasks under conditions of either high or low proactive interference. The performance of older participants was significantly poorer in the high-interference conditions, but when proactive interference was low, the performance of the older group no longer differed from that of the younger group. Thus the older participants were less able to inhibit intrusive interference.
Using fMRI neuroimaging, Gazzaley et al. (2005) showed that healthy older adults demonstrate a pronounced deficit in the inhibition of neural activation associated with task-irrelevant representations, while showing normal neural activation associated with the enhancement of taskrelevant representations. Gazzaley et al. also found that this lack of inhibitory control was correlated with impaired working memory performance. Thus it appears that one important executive function is to ensure that task-irrelevant information is ignored or suppressed, and evidence indicates that this inhibitory function can deteriorate with age.
Ageing can also affect executive attentional control in relation to updating in working memory; older participants perform relatively poorly on tasks that require the updating of information in working memory. This is important, because the successful completion of some cognitive tasks requires that the content of working memory is updated from its initial state. For example, a feature of many cognitive tasks is that initial goals are updated as the participant progresses through the task. Indeed, failure in the retention or updating of goals is a major source of working memory errors in older persons. Goal retention and updating requires executive attentional control, and the deterioration of such control in older people can produce ‘goal neglect’ in working memory tasks. It is worth noting that goal management and inhibitory control are related working memory functions, i.e. the former can assist the latter. Executive attentional control enables people to actively maintain and update goals and to use goal maintenance to inhibit contextually inappropriate responses; failures of executive attentional control can be regarded as instances of goal neglect (Braver and West, 2008).
CAN TRAINING IMPROVE WORKING MEMORY IN OLDER PEOPLE?
Although ageing can impair executive attentional control in working memory, recent evidence indicates that executive functioning and working memory in general can be improved by suitable training of older persons. Before looking at this evidence it is important to distinguish three different types of improvement in performance that may result from training. First, improvements in ‘target training’ refers to gains in performance of a task when that task is trained. Second, ‘near transfer’ gains refer to improvements in performance of tasks that were not trained but that are related to the same underlying cognitive ability or construct as the trained task. Third, ‘far transfer’ refers to improvements in the performance of tasks that were not trained and are not related to the same cognitive ability or construct. Gains confined to the target task can be of value, but the transferable improvements of near and far transfer may be of greater usefulness to older people in everyday cognition.
Heinzel et al. (2013) argue that far transfer requires training procedures that involve: executive control, the use of attention, processing speed or conscious cognitive control and that adaptive working memory training meets many of these requirements. Heinzel et al. chose the much-used ‘n-back’ working memory task because it appears to load the executive component of working memory and involves speeded responding. The n-back task has a number of differing versions and has been used in many studies of working memory. Essentially, it requires participants to register a series of random stimuli (typically numbers or letters) and compare the currently presented stimulus with their memory of previously presented stimuli in order to indicate when there is a match. The term ‘n-back’ refers to the number of serial positions backward that the participant is instructed to refer to (e.g. ‘three back’). Since each successive stimulus presentation effectively updates the identity of the stimulus occupying the target position, the participant must also update their memory of the target (see Figure 5.3). The participant must also inhibit the item previously occupying the target position.
Figure 5.3 An illustration of the n-back task where n = 3.
As discussed above, updating and inhibitory control are two executive control processes that make demands on the attentional resources of working memory. In Heinzel et al.’s study, participants were required to respond to only one feature of the stimuli (i.e. number identity), but in other studies participants have to retain and process two features (e.g. identity and spatial location), substantially increasing the level of task difficulty. Heinzel et al.’s results demonstrated that healthy older adults (aged 61–75 years) improved their performance on target training and produced significant far transfer effects on processing speed, episodic memory and near transfer effects of short-term memory. Heinzel et al. conclude that far transfer is possible in older persons and that working memory training can be effective for both old and young adults.
In a related study, Zinke et al. (2014) examined how much training was necessary to produce significant improvements in target training and transfer effects in a sample of eighty older adults aged 65–95 years. Using a short training programme consisting of only nine 30-minute sessions, they found consistent and substantial gains in target training on visuospatial, verbal and executive working memory tasks and far transfer to fluid intelligence. It was also found that the training gains were greater for the participants who showed lower scores on the baseline executive and working memory measures, indicating that lower-ability older adults gained most benefit from working memory training (Zinke et al., 2014).
It should be noted that key to the successful training of older people is the use of adaptivity. Adaptive training adjusts the level of difficulty of the training task to match the trainee’s current level of performance. For example, Richmond et al. (2011) demonstrated that training in spatial and verbal versions of an adaptive complex working memory span task produced clear near transfer and far transfer gains in 60–80-year-old participants. Richmond et al. argue that their complex working memory span training tasks involve ‘interweaving’ a storage task and an unrelated processing task that places heavy demands on general attentional control mechanisms, and that repetition strengthens these mechanisms to produce generalisable cognitive benefits. Interventions that fail to use adaptive training may produce weak or non-existent training effects.
It is important to point out that not all empirical studies provide support for the view that training can produce transfer gains. For example, Redick et al. (2013) found no evidence of an improvement in general cognitive ability in their study (i.e. no far transfer) and training gains were confined to target training. Redick et al. argue that those studies which have found positive transfer effects used no-contact (‘passive’) control groups rather than contact (‘active’) control groups. As a result, the experimental group may have been exposed to influences that the controls were not. However, it is worth noting that in fact at least 39 per cent of relevant experiments recently reviewed did use an active contact control group (see Karbach and Verhaeghen, 2014). For example, a study conducted by Brehmer et al. (2012) included a contact control group who undertook the same 5 weeks’ computerised working memory training programme as the experimental group. Both groups used the same training tasks and differed only in that the experimental group received adaptive training while the control group experienced an unvarying low level of task difficulty (i.e. non-adaptive training). Relative to the controls, older adults showed a significant improvement in the performance of both near and far transfer tasks and this improvement was maintained in a follow-up conducted 3 months later.
Thus it appears that training can produce transfer effects even when an active control group is used. Nevertheless, given the negative results obtained by some studies it is wise to consider the results of meta-analyses which have combined data from a number of different training studies conducted by differing researchers. Karbach and Verhaeghen (2014) examined data from sixty-one independent samples taken from forty-nine published papers using participants aged 63–87 years. The results demonstrated that working memory and executive functioning training each produce significant improvements in near and far transfer with moderate effect sizes (gains in target training were even greater). Moreover, these effects were present even when the analysis confined itself to studies that had used active contact controls. In terms of everyday cognition, the far transfer gains are of particular importance because they demonstrate a substantial improvement in fluid intelligence, suggesting that the effects of training generalise to tasks which are relevant to daily life (Karbach and Verhaeghen, 2014). Other recent meta-analyses have produced consistent results (e.g. Karr et al., 2014). It therefore appears that cognitive plasticity is retained in older age and that this plasticity can be exploited by interventions aimed at improving working memory functioning.
EVERYDAY ACTIVITIES THAT PROVIDE RESISTANCE TO COGNITIVE AGEING
In addition to formal training, there are everyday activities that also appear to provide ‘cognitive resistance’ to ageing among those who engage in them. For example, a number of studies have found that bilingual ism in older individuals is associated with improved working memory functioning. Bialystok et al. (2014) compared monolingual and bilingual older adults on a complex working memory task. Bilinguals showed greater resistance to interference in the performance of the task, and the difference between bilinguals and monolinguals was greater for the older participants than the younger participants. The results suggest that bilingualism in older persons is associated with greater executive control in working memory. Thus bilingualism may confer some ‘defence’ against the effects of ageing. Similarly, the playing of musical instruments appears to mitigate the effects of cognitive ageing. In Bugos et al. (2007), older adults aged 60–85 years received piano training. Participants and controls completed neuropsychological tasks at pre-training, post-training and at a 3-month follow-up. The piano training group exhibited significant improvements on a number of the tasks and the authors conclude that piano instruction may provide an effective intervention for cognitive ageing.
Finally, it is worth noting that many empirical studies have consistently demonstrated that gains in working memory and cognitive functioning generally are associated with regular aerobic exercise. Indeed, in a meta-analysis that included twenty-four studies of the effects of physical exercise on cognitive functioning in the elderly, Karr et al. (2014) found that overall there was a reliable and substantial effect of physical training on executive functions. This raises the question: by what mechanism does physical exercise enhance cognitive function in the old? This question was addressed by Colcombe et al. (2006) in a neuroimaging study in which older adults aged 60–79 years engaged in an aerobic exercise programme over 6 months. Prior to commencement of training, MRI brain scans were obtained and participants were scanned again at completion of the training programme. There were two control groups: one was a group of elderly participants who engaged in a programme of non-aerobic stretching and toning exercises (active control) and the other was a group of younger adults who did not engage in the exercise intervention (passive control). The MRI scan revealed a significant increase in brain volume in grey- and white-matter regions as a function of aerobic training in the older participants. The older control group who engaged in non-aerobic exercises showed no increases in brain volume, nor did the younger control group. The greatest increases in brain volume were evident in the frontal lobes in areas known to provide the neural substrate of executive functions in working memory. The authors suggest that this growth of brain tissue could result from increased levels of neurotrophic factor and insulin-like growth hormone which promote neuron proliferation and survival, the growth of brain capillaries and increased numbers of dendritic spines in neurons.
It should be noted that all the studies discussed above used healthy older adults. The efficacy of working memory training in abnormal ageing is less clear. However, recent evidence suggests that, at least in milder forms of impairment, cognitive training can be beneficial. For example, Gates et al. (2011) conducted a systematic review of ten studies involving a total of 305 participants diagnosed with mild cognitive impairment (MIC). Cognitive training was found to produce moderate-sized effects of training gains on memory performance and on global cognitive measures in a majority of studies; computer-based cognitive training produced the greatest effect sizes and enhanced generalisation of benefits, while memory strategy training produced less favourable results. Training gains have also been obtained in early-stage Alzheimer’s disease (AD) patients. For example, in Cipriani et al. (2006) AD patients undertook a computerised cognitive training programme which included exercises aimed at stimulating learning of new information and semantic memory. The AD participants showed gains on executive functions and general cognitive status as well as improvements in target training. However, the effectiveness of cognitive intervention in late-stage dementia is in some doubt and further work in this area is required.
Finally, we have seen that the efficacy of working memory training is based largely on improvements in executive functioning. In relation to the non-executive functions of working memory, recent work points to an interesting explanation of training gains. Although there is a substantial body of evidence demonstrating that older persons generally perform relatively poorly on non-executive working memory tasks, it is unclear whether this is due to poorer cognitive ability or to the use of sub-optimal task strategies. For example, Logie et al. (2015) point out that older participants tend to verbally recode visually presented material, resulting in poorer performance in visual memory tasks. Indeed, on the basis of their analysis of a number of relevant studies, Logie et al. conclude that the poorer performance of older persons on some working memory tasks reflects the use of inefficient task strategies rather than an age-related decline in the cognitive abilities that these tasks are designed to measure. If Logie et al.’s analysis is correct, it follows that the positive effects of training non-executive working memory may result from the development of more efficient task strategies rather than from training specific working memory resources. Thus training may enable dysfunctional strategies to be ‘unlearnt’ and replaced with more efficient approaches to task performance.
5.3 INDIVIDUAL DIFFERENCES IN WORKING MEMORY CAPACITY
This section will examine how individual differences in working memory affect everyday cognition. It is worth noting the link between this section and the preceding section in the sense that individuals with above-average working memory capacity may be more resistant to the effects of ageing than those with lower capacity. Across all adult ages a key predictor of individual differences in a wide range of cognitive skills is variation in working memory capacity (Engle, 2002). Randall Engle and his colleagues have identified a number of empirical studies which demonstrate a relationship between working memory capacity and performance in many everyday activities, including reading comprehension, speech comprehension, spelling, spatial navigation, learning vocabulary, note-taking, writing, reasoning and complex learning (Engle et al., 1999). Many studies have found that performance in these, and related tasks, can be predicted by individual differences in the working memory capacities of the participants. The measure of individual working memory capacity is known as ‘working memory span’ (WMS) and several tests of WMS have been devised (see Broadway and Engle, 2010; Daneman and Carpenter, 1980; Turner and Engle, 1989). All such tests involve storage and concurrent processing. For example, Daneman and Carpenter’s (1980) span test requires participants to read lists of sentences. In addition to processing the sentences for meaning (the processing load), the participants are also required to recall the last word in each sentence (the storage load). Turner and Engle (1989) developed an ‘operation span’ test in which participants are required to store words while processing simple arithmetic problems (see Figure 5.4). In both tests the participant’s working memory span is taken to be the number of words correctly recalled. However, performance in these tasks not only reflects individual differences in working memory capacity but also differences of ability in reading comprehension or arithmetic skill. Therefore, in order to obtain an accurate measure of individual variation in working memory capacity, researchers should combine operation span, reading span and spatial working memory tests. Such batteries of tests are time consuming to administer. However, Foster et al. (2015) have recently demonstrated that substantially reducing the number of blocks, and also trials within blocks, of working memory span tasks does not reduce their predictive validity.
Figure 5.4 An example of Turner and Engle’s (1989) operation span test.
A recent working memory capacity test is Broadway and Engle’s (2010) ‘running memory span’ task, which requires participants to recall the final n items from presented lists that are m + n items long. It follows that when m is greater than zero the list will be ‘partially recalled’ and when m = 0 such trials will be ‘wholly recalled’. The purpose of including whole recall trials was to discourage participants from adopting a strategy of ignoring the recently presented items. It was assumed that participants in the running memory span task may use an ‘active input processing’ strategy in which they rehearsed or grouped target items in advance of the test or they may use a ‘passive’ post-presentation strategy. Broadway and Engle designed their experiments so that they either supported or limited active input processing in order to isolate the two differing strategies from each other. Despite this, the results indicated that participants always used a passive, i.e. non-rehearsal, strategy regardless of condition. However, the findings did show very strong positive correlations between running memory span and performance on a range of working memory tasks, including operation span and reading span, thus validating running memory span as a measure of working memory capacity. On the basis of their results concerning the use of a passive strategy, Broadway and Engle conclude that contrary to a common assumption in the literature, the running memory span task does not rely on working memory updating but rather it taps in to the executive control of attention, attentional capacity and retrieval.
In addition to the measures of global working memory capacity developed by Engle and his colleagues, a number of studies have measured individual variation in specific components of working memory (e.g. Shah and Miyake, 1996). This work has revealed the orthogonality of modality-specific working memory resources (i.e. it is possible for an individual to score high on spatial working memory while scoring low on verbal working memory, and the converse). Moreover, this approach is not confined to laboratory-based studies; it has also been applied to real-world tasks such as way-finding in the built environment (e.g. Fenner et al., 2000) and learning computer programming languages (e.g. Shute, 1991; see Section 5.4).
5.4 WORKING MEMORY AND SOFTWARE DEVELOPMENT
LEARNING PROGRAMMING LANGUAGES
Software engineering is arguably the most challenging engineering discipline, partly because its inherent complexity places considerable demands on limited cognitive resources (Wang and Patel, 2009). Given that working memory limitations are a major source of error in skilled performance, working memory is a suitable construct for the purpose of assessing programming skills (Bergersen and Gustafsson, 2011). The importance of working memory in the acquisition of natural language (e.g. Gathercole and Baddeley, 1993) suggests that working memory may also play a role in learning computer programming languages, and a number of studies have examined this question (e.g. Bergersen and Gustafsson, 2011; Kyllonen, 1996).
Research in this area may have important educational applications: Shute (1991) argued that if we can identify the cognitive factors involved in the acquisition of programming skills, we may be able to ‘improve the design of effective computer programming curricula, providing educators with an explicit framework upon which to base instruction’ (p. 2). In Shute’s study, 260 participants received extensive instruction in the Pascal programming language. Following training, Pascal knowledge and skill were measured in three tests of increasing difficulty, each consisting of twelve problems. Test 1 required participants to identify errors in Pascal code. Test 2 involved the decomposition and arrangement of Pascal commands into a solution of a programming problem. Test 3 required participants to write entire programs as solutions to programming problems. Each participant also completed a battery of cognitive tests which examined working memory capacity, information-processing speed and general knowledge. The results revealed that ‘the working memory factor was the best predictor of Pascal programming skill acquisition [p = 0.001]. With all the other variables in the equation, this was the only one of the original cognitive factors that remained significant’ (p. 11).
Shute’s findings appear to have implications for teaching programming languages. Indeed, Shute concluded that the importance of working memory as a predictor of programming skill acquisition suggests that instruction should be varied as a function of individual differences in working memory span. There are a number of ways that this might be achieved. One approach might be to adjust the informational load during training so that it is commensurate with the working memory capacity of the trainee. Other techniques might involve supplying trainees with error feedback and the provision of external working storage to reduce the internal working memory load. In practice, it is likely that an effective approach would require that several such techniques were used in combination.
Shute interpreted her results as indicating that working memory contributes to both declarative and procedural learning in computer programming. Support for this view came from a study reported in Kyllonen (1996). In this work the performance of participants acquiring computer programming skill was examined in terms of orthogonal factors of procedural learning and declarative learning. Working memory capacity was found to account for 81 per cent of the variance in the declarative learning factor, while no other factor had a significant effect. Working memory capacity was also found to be the most influential determinant of procedural learning, accounting for 26 per cent of the variance on this factor. One interesting implication of these results is that the load placed on working memory by declarative information is greater than that imposed by the procedural content of the task. This may be because some procedures are partly automatised and consequently make less demand on working memory resources. It is worth noting that during training some of the initially declarative knowledge may become proceduralised, with the result that the load on working memory is reduced and resources are liberated for use on other components of the task.
EXPERT SOFTWARE DEVELOPMENT
The inherent cognitively demanding nature of software development means that software teams act as ‘cognitive agents’ (Chentouf, 2014). Given the centrality of working memory in human cognition it is likely that it plays a role in expert programming. Indeed, it may be possible to predict programming skill from individual differences in working memory span. This possibility was examined by Bergersen and Gustafsson (2011) in their study of sixty-five professional software developers. The programming skill of participants was assessed using a programming task instrument. The instrument contained twelve programming tasks in the Java programming language. In addition, the participants’ programming knowledge and experience of the Java programming language was assessed using a 30-item multiple choice test. All participants also completed three tests of working memory: operation span, symmetry span and reading span. The results showed that while working memory capacity did predict programming skill, the effect was mediated by programming knowledge. In other words, an individual with above-average working memory capacity but relatively poor programming knowledge would be unlikely to have above-average programming skills. These results demonstrate the importance of long-term memory in real-world working memory tasks. Indeed, there are few everyday working memory tasks that do not also involve declarative or procedural long-term knowledge. Working memory can be regarded as a ‘mental workbench’ where long-term knowledge is applied to process newly encoded material.
Another way in which long-term memory can interact with short-term is through ‘long-term working memory’. Altmann (2001) used a form of episodic long-term working memory or ‘near-term memory’ to model the behaviour of expert programmers engaged in the modification of large computer programs. Altmann grounds his model of near-term memory in the SOAR cognitive architecture (Newell, 1990). Altmann argues that during inspection of the program, the programmer is presented with considerably more information than can be retained in memory. During the session, the programmer will encounter items that relate to previously encountered details. The expert programmer will be able to retrieve these details by scrolling back through the listing to their location. Retrieval is accurate even when the number of such details exceeds working storage capacity. According to Altmann, this is because each time a detail is encountered, the programmer attempts to understand it by using their expert knowledge of programming. This produces an ‘event chunk’ specifying the episodic properties of the detail (e.g. its location in the listing), which are retained in near-term memory. Thus near-term memory provides a link between external information and expert semantic knowledge, with the result that key details can be retrieved when needed.
Recent research has attempted to find ways to reduce the ‘cognitive burden’ on software engineers. This has included designing artificial cognitive systems to support human software developers by carrying out some of the cognitive functions currently undertaken by human programmers (Chentouf, 2014).
5.5 WORKING MEMORY IN AIR TRAFFIC CONTROL
THE ROLE OF WORKING MEMORY IN THE AIR TRAFFIC CONTROL (ATC) TASK
The volume of air traffic has increased dramatically in recent years, and it is likely to increase further as air travel grows; if future demand is to be met safely, our understanding of the mental workload of the air traffic controller will need to improve (Loft et al., 2007). Several studies have identified working memory as playing a role in the performance of the ATC task (Smieszek et al., 2013). ATC is a complex and demanding safety-critical task and the air traffic controller deals with transient information to which a number of executive control processes must be applied. This information must be retained and updated in working storage for tactical use or strategic planning along with related outputs; as a result, performance of the ATC task is constrained by working memory limitations (Smieszek et al., 2013; Garland et al., 1999; Stein and Garland, 1993).
Working memory allows the controller to retain and integrate perceptual input (from the radar screen, flight strips and audio communications) while simultaneously processing that information to arrive at tactical and strategic decisions. Tactical information retained in working memory includes aircraft altitudes, airspeeds, headings, call-signs, aircraft models, weather information, runway conditions, the current air traffic situation and immediate and potential aircraft conflicts (Stein and Garland, 1993).
Figure 5.5 Air traffic controllers.
Source: copyright Angelo Giampiccolo / Shutterstock.com.
And since the air traffic situation is constantly changing, the contents of working memory must be constantly updated.
An overarching requirement of the en-route ATC task is to maintain ‘separation’ between aircraft (usually a minimum of 5 nautical miles horizontally). The controller must anticipate and avoid situations that result in a loss of separation (aircraft ‘conflicts’ or more generally ‘operational errors’). The dynamic nature of the air traffic environment ensures that this requires the execution of a number of control processes within working memory. One such control process involves the scheduling of actions. For example, a controller may instruct several aircraft within their sector to alter altitude or heading. It is imperative that these manoeuvres are carried out in an order that avoids the creation of conflicts between aircraft. In addition, scheduling must be responsive to unanticipated changes in the air traffic environment, which may require schedules to be updated (see Niessen et al., 1998; Niessen et al., 1999).
Dynamic scheduling of this sort is an important function of working memory (Engle et al., 1999). Another executive function of working memory is the capacity to process one stream of information while inhibiting others (Baddeley, 1996). Such selective processing is an important feature of the ATC task. For example, controllers reduce the general cognitive load of the task by focusing their attention on prioritised ‘focal’ aircraft (which signal potential conflicts) and temporarily ignore ‘extra-focal’ aircraft (Niessen et al., 1999). Moreover, dynamic prioritisation is itself an important control process in ATC that requires flexible executive resources.
It is worth noting that the flexible use of attentional resources is also regarded as key in many other dynamic tasks, including those relating to pilot cognition. Wickens and Alexander (2009) use the term ‘attentional tunnelling’ to describe the ‘allocation of attention to a particular channel of information, diagnostic hypothesis, or task goal, for a duration that is longer than optimal’. Wickens and Alexander provide the example of Eastern Airlines flight L1011 which crashed in the Florida Everglades. While focusing attention on what appeared to be a landing gear malfunction, the pilots failed to attend to their descending altitude, with tragic consequences. Similarly, many road traffic accidents in which drivers are speaking on mobile phones may involve attentional tunnelling on their conversations at the expense of attending to the driving task. It is possible that attentional tunnelling is also responsible for some of the operational errors in ATC, particularly under conditions of stress when ‘attentional narrowing’ may be present.
Clearly, ATC requires controllers to make use of a great deal of knowledge stored in long-term memory. During training, controllers acquire declarative and procedural knowledge without which they would be unable to perform the ATC task. Indeed, in ATC, working memory is dependent upon long-term memory for a number of key cognitive operations, including the organisation of information, decision making and planning (Stein and Garland, 1993). The temporary activation, maintenance and retrieval of information in long-term memory are processes controlled by the central executive component of working memory (Baddeley, 1996). Thus, working memory plays a key role in the utilisation of the long-term knowledge used to interpret and analyse information emanating from the air traffic environment.
The avoidance of air traffic conflicts is essentially a problem-solving task and problem resolution is a key information-processing cycle in ATC (Niessen et al., 1999). Working memory plays an important role in problem solving by retaining the initial problem information, intermediate solutions and goal states (Atwood and Polson, 1976). The working storage of goals and subgoals appears to be essential in a wide range of problem-solving tasks. Indeed, when the rehearsal of subgoals is interfered with, both errors and solution times increase (Altmann and Trafton, 1999). In ATC, goal management is a dynamic process because goal and subgoal priorities change as a function of changes in the air traffic environment. In executing a plan to attain a goal, the controller may need to retain in working storage a record of the steps currently completed and those that remain to be completed. Each time a step is completed, the contents of working memory need to be updated to record this fact.
Goals and subgoals can also change before they are attained. For example, changes in the air traffic situation can result in the removal or creation of goals and produce changes in the priority of existing goals. The management of goals is another important functional aspect of working memory and empirical studies have shown that when additional working memory resources are made available to goal management, problem-solving performance improves (e.g. Zhang and Norman, 1994).
The dynamic nature of the air traffic environment means that controllers must have an accurate awareness of the current and developing situation (Wickens, 2000). In this context the term ‘situation awareness’ refers to the present and future air traffic situation, and a number of studies have identified situation awareness as key to safe and efficient air traffic control (e.g. Endsley, 1997; Niessen et al., 1999). Experienced air traffic controllers often describe their mental model of the air traffic situation as the ‘picture’ (Whitfield and Jackson, 1982). The picture contains information about the fixed properties of the task and the task environment (e.g. operational standards, sector boundaries, procedural knowledge) as well as its dynamic properties (e.g. current and future spatial and temporal relations between aircraft). Thus, although some of the content of the picture is retrieved from long-term memory, working memory is involved in the retention of the assembled picture (Logie, 1993; Mogford, 1997). Moreover, the variable nature of the air traffic environment means that the picture needs to be repeatedly updated using executive control processes in working memory.
Endsley (1997) sought to identify and examine the psychological factors responsible for operational errors in en-route air traffic control. A total of twenty-five duty controllers observed re-creations of operational errors and reported on their situation awareness and cognitive workload. The results showed that under conditions of high subjective workload, situation awareness was compromised as attention was allocated to prioritised information. Endsley reports that under high workload, controllers had significant deficiencies in ongoing situation awareness, with low ability to report the presence of many aircraft, their locations or their parameters. When a high number of aircraft were present, controllers prioritised situation awareness of aircraft separation at the expense of other aspects of the situation.
These findings can be better understood by considering studies that have identified the detailed nature of situation awareness in ATC. Using a sample of experienced en-route controllers, Niessen et al. (1998, 1999) identified a number of ‘working memory elements’ (WMEs) that comprise the ‘picture’ used in ATC. They found that the picture consists of three classes of WMEs: objects, events and control elements. Examples of object WMEs are incoming aircraft, aircraft changing flight level and proximal aircraft. Events include potential conflicts of a chain or crossing kind. Control elements include selecting various sources of data (e.g. audio communication, flight level change tests, proximity tests), anticipation, conflict resolution, planning and action. Control procedures select the most important and urgent WMEs, which are arranged in working memory in terms of their priority. The continuously changing air traffic environment requires that ‘goal-stacking’ within working memory is a flexible process.
Clearly voice communication with pilots and other controllers is an important element of the air traffic control task. Via radio, the controller may convey instructions to pilots and receive voice communications from pilots. Voice communication errors can contribute to serious aviation incidents (Fowler, 1980). A tragic example is the collision between two 747s on the runway of Tenerife airport in 1977, which resulted in the deaths of 538 people and which was partly the result of a voice communication error (Wickens, 2000). Misunderstandings account for a substantial number of voice communication errors and many of these result from overloading working memory capacity (Morrow et al., 1993). Working memory assists speech comprehension by retaining the initial words of a sentence across the intervening words, thereby allowing syntactic and semantic analysis to be applied to the complete sentence (Baddeley and Wilson, 1988; Clark and Clarke, 1977).
In addition to comprehension failures, voice communication errors can also result from confusions between phonologically similar items in working memory. For example, the call-signs BDP4 and TCG4 contain phono logically confusable characters, increasing the risk of errors relative to phonologically distinct equivalents (Logie, 1993) producing a ‘phonological similarity effect’ (see Figure 5.6).
Figure 5.6 Examples of phonologically confusable letters and call-signs.
5.6 WORKING MEMORY AND MENTAL CALCULATION
THE ROLE OF WORKING MEMORY IN MENTAL ARITHMETIC
Working memory plays a key role in mental arithmetic (Hubber et al., 2014; Imbo and LeFevre, 2010; Vallée-Tourangeau et al., 2013). Exploring the role of working memory in this activity has applied relevance as mental calculation occurs in many real-world activities ranging from so-called ‘supermarket arithmetic’ to the technical skills used in employment and education (Hitch, 1978; Smyth et al., 1994). Why is working memory so important in mental arithmetic? In written arithmetic the printed page serves as a permanent external working store, but in mental arithmetic initial problem information and intermediate results need to be held in working memory (Hitch, 1978). In most individuals, mental calculation involving multi-digit numbers requires several mental operations rather than immediate retrieval of the solution. Working memory is used to monitor the calculation strategy and execute a changing succession of operations that register, retain and retrieve numerical data (Hunter, 1979). Intermediate results must be retained in working storage so that they may be combined with later results to arrive at a complete solution. Mental calculation is a task that involves storage and concurrent processing and is, therefore, likely to be dependent on working memory.
In an early study, Hitch (1978) demonstrated the involvement of working memory in mental calculation in an investigation of mental addition. Participants were aurally presented with multi-digit addition problems such as ‘434 + 81’ or ‘352 + 279’. The results showed that participants solve mental addition problems in a series of calculation stages, with the majority following a consistent strategy, e.g. ‘units, tens, hundreds’. More recent work has also shown that this ‘UTH’ strategy is more likely when problems involve carrying (e.g. Green et al., 2007). Working memory is used to retain the units and then the tens totals as partial results while the hundreds total is calculated. Hitch also found that solution times increased as a function of the number of carries required in the calculation and that carrying also loads working memory. In Experiments 2 and 3, Hitch found that effectively increasing the retention time for the ‘tens’ and ‘units’ totals resulted in the rapid decay of this information in working storage. Hitch concluded that in multi-digit mental addition, working memory is used to retain both initial material and interim results.
THE CONTRIBUTION OF WORKING MEMORY COMPONENTS
Since Hitch’s influential early work, a number of studies using a variety of approaches have also demonstrated the importance of working memory in mental arithmetic (e.g. Ashcraft and Kirk, 2001; Dumontheil and Klingberg, 2012; Hubber et al., 2014; Imbo and LeFevre, 2010; Logie and Baddeley, 1987; McClean and Hitch, 1999).
Several studies have attempted to identify the role of the different components of working memory in arithmetic. For example, Trbovich and LeFevre (2003) asked ninety-six adult participants to solve multi-digit arithmetic problems. In one condition a visual memory load was also present; in another condition a phonological load was present. The presentation format was also manipulated: problems were presented in either a vertical columnar format or a horizontal row format (see Figure 5.7). When the problems were presented horizontally, performance was impaired by the phonological load, but when problems were presented in the vertical format, the visual load caused the greatest impairment. Trbovich and LeFevre conclude that both visual and phonological working memory are involved in mental arithmetic, but the contribution that each makes is dependent on the presentation format of the problem material.
Figure 5.7 Examples of problem presentation formats in Trbovich and LeFevre (2003). Upper row: horizontal presentation format; lower row: vertical presentation format.
In a more recent study, Imbo and LeFevre (2010) also manipulated problem complexity in the form of carrying or borrowing operations along with presentation format. Again, either visual or phonological load impaired performance in solving either subtraction or multiplication problems. However, while these effects were present in both Canadian and Chinese participants, only the Chinese showed effects of problem complexity and presentation format, suggesting that cultural differences can be influential.
Using a different approach, Dark and Benbow (1991) examined the working memory representational capacity of participants who scored highly on either mathematical ability or verbal ability. The results showed enhanced capacity for numerical information for the high mathematical group and enhanced capacity for words for the high verbal group. Moreover, the high mathematical ability group were found to be more efficient at representing numbers in the visuospatial sketchpad. Indeed, several studies point to the importance of visuospatial working memory in mental calculation. Ashcraft (1995) argues that in mental arithmetic the visuospatial sketchpad is used to retain the visual characteristics of the problem as well as positional information. This is evidenced by the fact that participants frequently ‘finger write’ mental calculation problems in the conventional format (see also Hope and Sherrill, 1987). Visuospatial working memory makes a contribution to any mental arithmetic problem that ‘involves column-wise, position information’ and ‘to the carry operation, given that column-wise position information is necessary for accurate carrying’ (Ashcraft, 1995, p. 17; see also Trbovich and LeFevre, 2003). Converging evidence for the involvement of visuospatial working memory in arithmetic has come from neuroimaging studies. Using fMRI, Dumontheil and Klingberg (2012) found that greater activation in the left intraparietal sulcus area of the brain is associated with poorer arithmetical performance in participants when assessed 2 years later. It is worth noting that compared with the use of behavioural predictors alone, the addition of brain-imaging data improved the predictive accuracy of assessments by 100 per cent.
While the visuospatial sketchpad appears to have an important role in mental calculation, it is unlikely to operate in isolation. Indeed, Ashcraft (1995) regards the phonological loop as also contributing by retaining the partial results generated during mental arithmetic. Consistent with this, McClean and Hitch (1999) asked participants to complete a battery of working memory tests measuring performance dependent on either the phonological loop, visuospatial working memory or the central executive. A comparison was made between participants with poor arithmetic ability and those with normal arithmetic ability. The results revealed that while the groups failed to differ on phonological loop tests, their performance was significantly different in tests of spatial working memory and central executive functioning. McClean and Hitch concluded that spatial working memory and executive functioning appear to be important factors in arithmetical attainment. These results are consistent with studies that demonstrate the importance of visualspatial ability in the arithmetic performance of adults (e.g. Morris and Walter, 1991). Heathcote (1994) found that the phonological loop was responsible for the retention of partial results and contributed to the working storage of initial problem information. Heathcote’s results suggested that the phonological loop operates in parallel with the visuospatial sketchpad, which retains carry information and provides a visuospatial representation of the problem. Operating in isolation, the capacity of the phonological loop may be largely depleted by the requirement to retain material in calculations involving three-digit numbers. The independent capacity of visuospatial working memory may be used to support phonological storage. It is worth noting that the capacity of visuospatial working memory for numerals is greater than the capacity of phonological working memory for their verbal equivalents (Chincotta et al., 1999). Fuerst and Hitch (2000) found that mental addition was impaired by concurrent articulatory suppression (i.e. repeated vocalisation of an irrelevant word), a task known to load the phonological loop. When the problem information was made continuously available for inspection, articulatory suppression ceased to influence performance. These results support the view that the phonological loop is involved in the retention of the initial problem material.
The importance of the phonological loop was also demonstrated in Logie et al.’s (1994) study of mental calculation. In their experiments, participants were required to mentally add two-digit numbers presented either visually or auditorily. Performance was disrupted by concurrent articulatory suppression. The results suggested that subvocal rehearsal assists in the retention of interim results (i.e. running totals), as found in previous studies (e.g. Heathcote, 1994; Hitch, 1980; Logie and Baddeley, 1987). Logie et al. also found that a concurrent spatial task impaired performance on visually presented problems, again suggesting that the phonological loop and the visuospatial sketchpad can both play a role in mental calculation.
Recent work has explored the role of the central executive in mental arithmetic. Hubber et al. (2014) found that while maintaining visuospatial information in the visuospatial sketchpad plays a small role in solving addition problems, the central executive makes the greatest contribution to performance. Indeed, a key finding of Logie et al.’s study was that the greatest impairment of mental calculation was produced by a random generation task known to load the central executive. This result is consistent with the view that the central executive is involved in the retrieval and execution of arithmetical facts and strategies stored in long-term memory (Heathcote, 1994; Hitch, 1978; Lemaire et al., 1996). Clearly, mental calculation would not be possible without the utilisation of long-term knowledge relevant to the task. The central executive appears to have a role in the retrieval and implementation of procedural and declarative arithmetical knowledge. An example of essential declarative knowledge is that mental calculation is dependent upon access to numerical equivalents (i.e. arithmetical facts) such as 7 · 7 = 49 or 8 + 4 = 12. Mental arithmetic also requires procedural knowledge about calculative algorithms, e.g. the rule to follow when required to apply the operator ‘·’ to two numbers. Having retrieved the appropriate algorithm, the central executive then applies that rule and monitors and updates the current step in the procedure. Thus, the executive is responsible for the execution of essential calculative operations, e.g. the execution of carry operations (Fuerst and Hitch, 2000).
MULTIPLE WORKING MEMORY COMPONENTS IN NUMERICAL PROCESSING
Mental calculation appears to require both verbal and visuospatial working memory subsystems (Trbovich and LeFevre, 2003) together with considerable executive resources (Hubber et al., 2014). This is not entirely surprising given that mental arithmetic can be a task that places heavy demands on cognitive resources. Collectively, the findings discussed above can be explained by the ‘triple code model’ of numerical processing proposed by Dehaene and his colleagues (Dehaene, 1992; Dehaene et al., 1993; Dehaene and Cohen, 1995). In this model, during multi-digit mental arithmetic, numbers are mentally represented in three different codes. First, a visual Arabic form in a spatially extended representational medium (e.g. ‘592’); in this code ‘numbers are expressed as strings of digits on an internal visuo-spatial scratchpad’ (Dehaene and Cohen, 1995, p. 85); second, a verbal code which is linked to phonological representations; third, an analogical spatial representation which expresses the magnitude of numbers and contributes to approximate solutions. During complex mental calculation, all three codes operate in parallel because there is a permanent transcoding back and forth between the visual, verbal and analogical representations (see Figure 5.8). Visuo spatial working memory is involved in the representation of the visual code and the analogical magnitude code. The phonological loop retains information coded in a verbal format.
Figure 5.8 A simplified representation of Dehaene’s ‘triple code model’ of numerical processing.
More recently, Dehaene’s triple code model has been validated by neuroimaging studies. Using fMRI, Schmithorst and Brown (2004) examined brain activity in neurologically intact adult participants engaging in the complex addition of fractions. Independent task-related components were found with activation in the bilateral inferior parietal, left perisylvian and ventral occipitotemporal areas corresponding to the three distinct functional neuroarchitectures of the triple code. Activity in the bilateral inferior parietal area is associated with abstract representations of numerical quantity and is therefore linked to Dehaene’s analogical magnitude code used to access semantic knowledge about the relative positions of fractions on a mental number line. The left perisylvian network, which includes Broca’s and Wernicke’s areas, is associated with language functions and the authors suggest that activity here provides the neural basis of the verbal code used for retrieving declarative arithmetic facts and conversions to common denominators. Finally, the ventral occipitotemporal region is associated with the ventral visual pathway and may have been used to process the visual Arabic code, including spatially manipulating numerals used in fraction additions (Schmithorst and Brown, 2004). Interestingly, the authors also found some activation in the prefrontal area, which as we’ve seen is associated with executive functions in working memory. Consistent with this, the authors argue that their findings support Dehaene’s proposal that these areas are involved with coordinating the sequencing of processing through the triple code modules in the appropriate order and holding intermediate results in working memory.
WORKING MEMORY AND MATHEMATICS ANXIETY
The working memory model may also provide a useful framework to explore the relationship between emotion and cognition. An example is the association between working memory and mathematics anxiety. Vallée-Tourangeau et al. (2013) point out that mathematics anxiety impairs performance in mental arithmetic tasks. In a ‘static’ condition, participants were asked to find solutions to problems without using their hands. This was compared with performance in an ‘interactive’ condition where problems were presented in the form of manipulable tokens. The results showed that levels of maths anxiety were only correlated with arithmetic performance in the static condition; in the interactive condition maths anxiety ceased to be influential. A mediation analysis indicated that the effect of maths anxiety on arithmetic performance was mediated by working memory capacity in the static condition but not in the interactive condition. Vallée-Tourangeau et al. argue that interactivity promotes the combining of cognitive resources with external resources so as to augment working memory capacity.
Ashcraft and Kirk (2001) found that the calculation-based working memory span of participants was reduced by mathematics anxiety. The reduction in working memory capacity caused by maths anxiety severely impaired performance on mental addition problems. The diminution of working memory span was found to be temporary, the result of online processing of specifically maths-related problems. Mathematics anxiety appears to impair the efficiency of the central executive in executing procedural operations such as carrying, sequencing and keeping track in multi-step problems. Ashcraft and Kirk conclude that their results directly confirm and extend Eysenck and Calvo’s (1992) ‘processing efficiency theory’, which predicts that the intrusive thoughts and worries associated with anxiety compete with cognitive tasks for limited working memory resources. Anxiety produces a reduction in executive processing capacity by compromising selection mechanisms, allowing intrusive thoughts and distracting information to compete for limited processing capacity. An important implication of these findings is that interventions aimed at reducing maths anxiety may produce substantial improvements in the performance of mathematics-related tasks. Moreover, the performance of individuals with below-average general working memory capacity is likely to be particularly sensitive to further decrements in capacity produced by anxiety. Therefore it is this group who are likely to benefit most from anxiety-reducing techniques.
5.7 WORKING MEMORY AND HUMAN–COMPUTER INTERACTION (HCI)
WORKING MEMORY ERRORS IN HCI
Interaction with digital devices is now a ubiquitous everyday occurrence. The field of human–computer interaction is not confined to interactions with desktop or laptop computers; HCI research encompasses the use of many different forms of information technology (Dix et al., 1997). HCI is wide ranging and not always immediately obvious. For example, the use of automatic teller machines (ATMs) to make cash withdrawals is an instance of everyday HCI. Byrne and Bovair (1997) studied the cognitive errors associated with using ATMs to make cash withdrawals from bank accounts. This study examined a type of systematic error known as a ‘post-completion error’. Post-completion errors occurred when the user completed their task of withdrawing cash but failed to remove their bank card from the ATM. In general, post-completion errors tend to happen when users have an additional step to perform after their primary goal has been attained. Byrne and Bovair found that this form of error only occurred when the load on working memory was high. In these circumstances the presence of an additional step overloads limited working memory capacity. It is worth noting that the occurrence of post-completion errors led to the redesign of ATMs, with the result that they now only dispense cash after first returning the card to the user.
Clearly cognitive load is a key factor in HCI and well-designed user interfaces may seek to minimise demands on working memory with the aim of avoiding user errors. Cognitive load in HCI can be measured in a range of ways, including performance measures, self-report and neurophysiological measures. An example of the latter comes from Gevins et al. (1998), who assessed working memory load in HCI tasks by monitoring electrical activity in the brains of users using electroencephalography (EEG). Theta activity in the frontal lobes increased and alpha activity decreased as the working memory load of the task increased.
Because the cognitive limitations of users can contribute to error production in human–machine tasks, some HCI researchers have pointed out that the design of user interfaces should take account of individual differences in working memory capacity. One interesting approach to this is ‘adaptive interface design’ in which the system assesses the user’s cognitive profile and adjusts its interface to match the limitations, abilities and interests of the user (e.g. Jerrams-Smith, 2000). Empirical studies have shown that adaptive interfaces can minimise error, encourage learning and have a positive effect on the user experience (Jerrams-Smith et al., 1999). The concept of adaptivity influenced Gwizdka and Chignell (2004) in their investigation of how individual differences in working memory capacity affect the everyday activity of receiving and responding to email. Received email often refers to ‘pending tasks’, i.e. tasks that must be completed in the future. Many users allow such emails to accumulate in their inboxes and then later go through their received mails in order to identify the pending tasks. Gwizdka and Chignell argue that this can be a demanding and error-prone activity for users with lower cognitive ability. They compared the performance of users assessed as high or low working memory span in a task that required users to find information relating to pending tasks in email messages. The results showed that users with low working memory capacity performed relatively poorly in a header-based task when a visual ‘taskview’ interface was used but that there was no difference between high and low working memory users when an MS Outlook interface was used.
The ubiquity of working memory errors may also have implications for the design of telephonic communication interfaces. A common form of telephone-based interaction involves the selection of menu items in automated systems. Huguenard et al. (1997) examined the role of working memory in phone-based interaction (PBI) errors when using such a system. Guidelines for the design of telephone interfaces emphasise the importance of not overloading the user’s short-term memory capacity. In particular, these guidelines advocate the use of ‘deep menu hierarchies’, which limit menu structures to a maximum of three options per menu. However, Huguenard et al.’s results indicated that deep menu hierarchies do not in fact reduce PBI errors. This is because although deep menu hierarchies reduce the storage load, they increase the concurrent processing load in working memory. In addition to its obvious practical implications, this study demonstrates how the working memory concept can provide a more accurate prediction of empirical findings than approaches that view temporary memory entirely in terms of its storage function.
OLDER COMPUTER USERS
Given the importance of working memory in HCI it appears that factors that result in a diminution of working memory capacity may have a detrimental effect on the performance of computer-based tasks. As we saw earlier, normal ageing seems to produce a decline in the processing capacity of working nemory (Baddeley, 1986; Craik et al., 1995; Salthouse, 2015; Salthouse and Babcock, 1991) and a number of studies have examined the impact of age-related working memory decrements on performance in HCI tasks (e.g. Freudenthal, 2001; Howard and Howard, 1997; Jerrams-Smith et al., 1999). Indeed, the load imposed on working memory by computer-based tasks may be a particularly influential factor in the usability of interfaces for the elderly. This may have important implications for the design of interfaces that seek to encourage the elderly to take advantage of computer technology.
In pursuing this aim, Jerrams-Smith et al. (1999) investigated whether age-related decrements in working memory span could account for poor performance in two common tasks associated with the use of a computer interface. The results demonstrated that relative to younger adults, older participants performed poorly in a multiple windows task that involved working storage and concurrent processing. Older participants were also found to have smaller working memory spans than younger participants. In addition, the study examined the short-term retention of icon labels in the presence of a concurrent processing load. The results showed that under these conditions the ‘icon span’ of younger participants was greater than that of the seniors. It was concluded that interfaces that place a considerable load on working memory (i.e. those that require considerable working storage and concurrent processing) are unsuitable for many older users. Interfaces designed for seniors should enable users to achieve their task goals with the minimum concurrent processing demands. Such interfaces would require sequential rather than parallel cognitive operations, thereby reducing the load on working memory.
• Many everyday cognitive tasks require the temporary retention and concurrent processing of material. Psychologists use the term ‘working memory’ to refer to the system responsible for these functions.
• Working memory allows long-term procedural and declarative knowledge relating to task-specific skills to be applied to, and combined with, immediate input.
• The executive functions of working memory are particularly sensitive to the effects of ageing and recent research has examined interventions aimed at minimising or reversing cognitive ageing.
• In mental calculation the executive component of working memory applies knowledge of calculative algorithms and numerical equivalents to the initial problem information and interim results retained in working storage.
• Dynamic task environments such as air traffic control place heavy demands on working memory resources. ATC presents an everchanging task environment that requires dynamic scheduling of operations and the retention, prioritisation and updating of task goals.
• Working memory errors can occur in human–computer interaction; such errors can be minimised by the use of adaptive user interfaces that adjust to individual differences in the cognitive capacities of users.
• Alloway, T.P. and Alloway, R.G. (2013). Working memory: The connected intelligence. Hove: Psychology Press.
• Baddeley, A.D. (2007). Working memory, thought and action. New York: Oxford University Press.
• Logie, R.H. and Morris, R.G. (2015). Working memory and ageing. Hove: Psychology Press.