A Critical Evaluation of Baddeley and Hitch’s Working Memory Models

A Critical Evaluation of Baddeley and Hitch’s (1974) and Baddeley’s (2000) Working Memory Model

Early memory-models explained memory as a rather linear process of receiving sensory information, passing it on to short-term memory and later to long-term memory (see Atkinson & Shiffrin, 1968). To widen explanations of memory, Baddeley and Hitch (1974) proposed their working memory model (WMM) offering an interacting link between sensory storages and long-term memory. Working memory (WM) can be defined as control-system with limited processing and storing-capacity, and WMM consists of phonological-loop (PL), visuo-spatial sketchpad (VSS), central-executive (CE), and episodic-buffer (EB) (Baddeley, 2000). In the following, these WMM-subsystems will be shortly described to then present and discuss evidence for each subsystem. Finally, the overall model will be evaluated, and practical applications presented.

The PL (‘inner voice’) holds speech-based information (Baddeley, 2000). A study by Baddeley, Gathercole, and Papagno (1998) suggested PL serving to acquire new words, like foreign language, but not remembering familiar words. To evidence PL as individual subsystem, experiments repeatedly used both phonological similarities effect (PSE) and word-length effect (WLE). While PSE is the finding that serial recall accuracy decreases the more similar items sound (Larsen, Baddeley, & Andrade, 2000), WLE defines that shorter words produce better correct-order-recall, independent of word-set size (Guitard et al., 2018). Problematically, recent evidence suggests not only phonological but also semantic processes to be involved in PSE and therefore the WWM to be underspecified (Nishiyama, 2018; Norris et al., 2018; Schweppe, Grice, & Rummer, 2011a; Schweppe et al., 2011b). Furthermore, it remains unclear whether acoustic or articulatory similarities underlay PSE (Eysenck & Keane, 2015). Despite a large basis of evidence for WLE (see Baddeley, 2012), WLE’s underlying processes and relevance to support PL as individual subsystem remain controversial. Jalbert and colleagues (2011) argue for WLE simply being an effect of orthographic neighbourhood (there are more similar short than long words), and McGill (2019) found a reversed WLE. As WMM suggests PL as ‘inner voice’, action control would consequently need to be impaired by receiving parallel phonological input (Eysenck & Keane, 2015). That hypothesis could be confirmed in different dual-task experiments showing that articulatory suppression reduced action-control and error-rates increased when demands on action-control were high (Covre et al., 2018; Jaroslawska, Gathercole, & Holmes, 2018; Nishiyama, 2018; Tullett & Inzlicht, 2010). Furthermore, Schulze, Vargha-Khadem, and Mishkin (2018b), found a mutation of the FOXP2 gene to cause lower performance in PL but not central-executive or visuospatial-sketchpad, suggesting PL as a genetically influenced independent sub-system. However, mutations of FOXP2 were also generally associated with speech and language disorders (Argyropoulos et al., 2018; Onnis, Truzzi, & Ma, 2018) and might therefore not exclusively explain PL as individual subsystem. Overall, WMM does not sufficiently describe different information processed by PL and needs further specification to allow follow-up research.

The visuospatial-sketchpad (VSS) or ‘inner eye’ consists of two parts: visual cache and inner scribe, establishing, often combined, visual and spatial coding or processing of visual semantics (Baddeley, 2000). A variety of dual-task-studies on VSS evidenced the dissociation between visual and spatial memory (Darling, Della Sala, & Logie, 2007; Fallon et al., 2019; Logie & Pearson, 1997; Schulze, Koelsch, & Williamson, 2018a; Smith & Jonides, 1997). However, methodologies of these experiments are discussed and inconsistently used producing incongruent results. For example, the required tasks-difficulty remains controversial as more difficult tasks seem to produce a general attention-based interference and easier tasks interference-effects specific to the interference (see experiments by Klauer and Zhao (2004), compared to Vergauwe, Barrouillet, and Camos (2009)). After reviewing the literature on verbal and visuospatial forward and backward recall, Donolato, Giofrè, and Mammarella (2017) argue that WMM is not clearly supported by experimental evidence. On the other hand, activations of different brain areas during spatial and visual processing would suggest the differentiation of VC and IS. Different reviews of literature show specifically contributions of frontal lobes to spatial WM (Cona & Scarpazza, 2019; Owen et al., 2005; Rottschy et al., 2012). For visual WM, Cai et al. (2018) could identify frontal and parietal cortex as regions of activation and the importance of context of stimuli in visual WM processing. According to Zimmer (2008), most experiments found bilateral activations in visuospatial tasks showing activations in occipital and temporal lobes during spatial processing and parietal lobes during visual processing. Controversially, Cona and Scarpazza (2019) could not find clear bilateral activations in their review and concluded that the attribution of tasks/stimuli to specific prefrontal cortical areas remains difficult. Accordingly, more research is needed to show whether there are specific brain areas consistently activated in different visuospatial tasks, to further understand how processing and information of visual cache and inner scribe are combined and integrated, and how to apply different methodologies to measure features of these subsystems rigorously.

Baddeley (2000) described the central-executive as controlling instance interacting with all WMM-subsystems to monitor information, as well as to allocate attention and resources to different WMM-subsystems. Hence, problems with CE would cause difficulties in switching between different WMM-components (Baddeley, 2000). Recent evidence could support central-executive’s role controlling attention and task switching (Baddeley, 2012), but its role in higher cognition remains discussed (Schulze et al., 2018a). Covre and colleagues (2018) concluded that CE’s limited resources were responsible for decreased performance in multi-tasking instead of limited capacity of other WMM subsystems. Similarly, Katus and Eimer (2019) conducted an EEG study showing that bimodal dual-task-costs resulted not from WM subsystem-competition for shared storage capacity but limitations in CE’s coordinating of multiple cognitive processes in dual-tasks, giving a variety of implications for applied settings. For example, recent fMRI evidence showed that CE might play a major role in stuttering. Yang and colleagues (2019) found abnormal activations independent of memory-load, compensatory mechanisms in the right hemisphere for rehearsal-process-deficits and neural disconnections related to CE. Problematically, both nature and frequency of processes involving CE remain unclear as those are not clarified in WMM and experiments observing CE are facing the task-impurity-problem making it difficult to identify CE’s exact contributions (e.g., Eysenck & Keane, 2015). Therefore, The Switching, Inhibition, and Updating Model of Executive Function became an alternative, well-evidenced and popular model among cognitive psychologists (Jewsbury, Bowden, & Strauss, 2016).

The most recently added subsystem is the episodic-buffer (EB), comprising a system allowing temporary information-storage from WM or long-term-memory as unitary episodic representation (Baddeley, 2000). EB’s information-storage and information-integration are hypothesised to be needed as CE’s storage-capacity is limited. Experiments by Baddeley et al. (2009) and Allen et al. (2012) showed an effect of the amount of input on task-performance independent of the tasks-range suggesting not CE to be responsible, but EB as a separate storing component. Problematically, WMM does not specify how information from phonological-loop and visuospatial-sketchpad are combined to shape uniform representations in EB. WMM also suggests other sensory information, such as smells and tastes, to be stored in EB, but relevant evidence is still lacking (Eysenck & Keane, 2015).

Overall, WMM has stood robust over the last 40 years in a fast-changing field due to its ability to explain all complex cognitive tasks by including active processing and transient information-storage, and several research-based refinements (e.g., Baddeley, 2000). Accordingly, WMM became one of the most important memory-models not only to help explaining memory processes theoretically but also practically (Andrade, 2001). For example, meta-analyses showed that traumatic brain-injury predicted lower WM capacity (Dunning, Westgate, & Adlam, 2016; Rowley et al., 2017) but that training could sustainably improve WM in brain-injury patients with medium-to-large effect sizes (Weicker, Villringer, & Thöne-Otto, 2016). Furthermore, lower WM capacity could be associated with anxiety (Moran, 2016), or ADHD (Kofler et al., 2018); for example, because of inhibition-deficits (Alderson et al., 2017). Moreover, measures of WM are used as measures of cognitive load in different contexts (e.g., education: Anmarkrud, Andresen, & Bråten, 2019). WM also relates to language development. Recent meta-analyses by Peng et al. (2018) and Kudo, Lussier, and Swanson (2015) found a relationship between (specifically verbal) WM and reading. Other studies showed relationships between developmental language-difficulties and WMM-subsystems (Henry & Botting, 2017), for example, developmental-language-disorder and CE (Pauls & Archibald, 2016).

[...]