Previous neuroimaging studies have identified brain regions that underlie verbal working memory in humans. According to these studies a phonological store is located in the left inferior parietal cortex, and a complementary subvocal rehearsal mechanism is implemented by mostly left-hemispheric speech areas. In the present functional magnetic resonance imaging study, classical interfering and non-interfering dual-task situations were used to investigate further the neural correlates of verbal working memory. Verbal working memory performance under non-interfering conditions activated Broca's area, the left premotor cortex, the cortex along the left intraparietal sulcus and the right cerebellum, thus replicating the results from previous studies. By contrast, no significant memory- related activation was found in these areas when silent articulatory suppression prevented the subjects from rehearsal. Instead, this non-articulatory maintenance of phonological information was associated with enhanced activity in several other, particularly anterior prefrontal and inferior parietal, brain areas. These results suggest that phonological storage may be a function of a complex prefronto-parietal network, and not localized in only one, parietal brain region. Further possible implications for the functional organization of human working memory are discussed.
Working memory has been defined as a set of linked and interacting information-processing components that allows temporary storage and simultaneous manipulation of information in the brain, a function critical for higher cognitive functions such as language, planning and problem-solving (Baddeley, 1992). Among the models of human working memory, the one provided by Baddeley and Hitch has been the most influential during the last two decades (Baddeley and Hitch, 1974). It divided short-term memory, which formerly had been considered as an unitary system, into three principal components: the ‘central’ executive as a complex attentional- controlling system and two subsidiary ‘slave’ systems designed to maintain representations of information of different modalities, i.e. the visuospatial sketchpad and the phonological loop. The phonological loop is considered to represent the neural correlates of verbal short-term memory in humans: a ‘passive’ storage component located in the left-hemispheric posterior parietal cortex and an ‘active’ subvocal rehearsal mechanism implemented by mostly left-hemispheric speech areas that include Broca's area, lateral and medial premotor cortices and, additionally, the contralateral cerebellum (Paulesu et al., 1993; Awh et al., 1996). The visuospatial sketchpad, on the other hand, is the hypothetical working memory component that is specialized for the processing and storage of visual and spatial material. Whether this component can be further subdivided into systems for spatial and visual object information (Gathercole, 1994; Hecker and Mapperson, 1997) and, in close analogy to the functional architecture of the phonological loop, into ‘passive’ storage and ‘active’ rehearsal mechanisms (Washburn and Astur, 1998; Awh et al., 1999), is still debated.
The aim of the present investigations was to reinvestigate the functional neuroanatomy of verbal working memory by measuring brain activity during verbal short-term memory tasks under varying task conditions and, in particular, by looking into activity changes associated with the well-established behavioral effect of articulatory suppression. This effect refers to the observation that verbal short-term memory is reduced when one has to perform other concurrent articulations, and presumably is caused by a disruption of the rehearsal mechanism of the phonological loop (Murray, 1968; Baddeley et al., 1984). Thus, according to the proposed two-component architecture of the phonological loop (Baddeley et al., 1984), one could expect that when articulatory suppression is performed at an articulation rate that does not permit any rehearsal of the information to be remembered, verbal working memory performance would have to rely on the second component, i.e. the phonological store, only.
In order to avoid motion artifacts in the functional magnetic resonance imaging (fMRI) scans that may result from overt articulations, silent articulatory suppression was chosen as an appropriate experimental paradigm that has already been shown to induce similar behavioral effects as overt articulatory suppression (Wilding and White, 1985). This experimental approach was validated in a separate behavioral experiment that directly compared the effects of overt and silent articulatory suppression with each other.
Materials and Methods
Participants were 16 consistent right-handers (eight men and eight women; mean age: 23.6 ± 3.4 years) who gave written informed consent.
The subjects were seated ~65 cm from a computer monitor. The stimulation protocol was the same as in the fMRI experiment described below. Overall, 54 memory trials were presented for each secondary task condition (single task, dual task with alternating finger tapping and dual task with articulatory suppression). From the memory trials under articulatory suppression, either the first or the second half (27 trials each) was performed with silent or overt counting, respectively, with the order randomized across subjects.
Participants were 11 consistent right-handers (six men and five women; mean age: 24.7 ± 4.3 years). The project was approved by the regional ethical committee, and all subjects gave written informed consent.
Experimental Design and Data Acquisition
A 3.0 T MRI scanner (Bruker Medspec 30/100) with a circularly polarized head coil was used to obtain a high-resolution structural scan for each subject followed by three runs of 518 gradient echo-planar image (EPI) volumes each (TR = 2 s, TE = 40 ms, flip angle = 90°; number of slices = 16, voxel size = 3 × 3 × 5 mm3, distance factor = 0.2) that were synchronized with stimulus presentation by means of ERTS (Experimental Run Time System, Version 3.11, BeriSoft Cooperation, Frankfurt am Main, Germany). The pretrained subjects performed verbal item-recognition tasks (Sternberg, 1966; Awh et al., 1996) in alternation with letter case judgment tasks. Each experimental trial began with a 1 s presentation of four letters, which were randomly taken out of a set of eight (in German) phonologically similar letters, followed by a 4 s fixation delay, and then a 1 s presentation of a single letter (see Fig. 1). Trials were separated by a 1 s fixation period. A cue instructed the subjects to either quickly read and memorize the four target letters, maintain them during the delay and to decide whether the probe letter matched one of these items or not (left/right button press with the index/middle finger of the right hand) or, alternatively, to read them without memorizing and maintaining, and to judge whether the single letter was uppercase (left button) or lowercase (right button). The matching proportion was pseudorandomized to 50% in each condition. Each of these two task variants (verbal item recognition and letter case judgment tasks, the latter serving as control conditions, which were matched to the corresponding memory task with respect to visual, motor and unspecific cognitive processing) lasted for 22 s (three trials per 7 s plus a 1 s cue presentation), and they were systematically alternated, as depicted in Figure 1, resulting in 154 s blocks. Different blocks were arranged in a counterbalanced order and varied with respect to the 4 s delays, which were either unfilled (single-task condition) or filled with tones that paced either silent counting (repeatedly from one to four; articulatory suppression) or alternating tapping (with the index and the middle finger of the left hand on response buttons; alternative dual-task condition). Thus, a 2 × 2 factorial design was employed with one factor being the short-term memory demands during the verbal item-recognition task and the other formed by a concurrent dual-task component that was achieved by either silent articulatory suppression or alternating finger tapping (Fig. 1). This latter, alternative dual-task condition appears comparable to articulatory suppression in terms of its general attentional demands and was introduced as a further control in order to differentiate the specific interference effect of articulatory suppression from possible more general dual-task effects. The stimulusonset asynchrony (SOA) of the 4000 Hz tones that paced either finger tapping or silent counting was 300 ms. During articulatory suppression, this pace ensured high demands on subvocalization processes, which successfully interfered with verbal rehearsal of the letters. In order to avoid confounding shifts of priorities in the different dual-task situations, the subjects were instructed to consider both silent counting and alternating finger tapping as the respective primary tasks and the item recognition and letter case judgment tasks as secondary. Furthermore, they were explicitly instructed to rehearse the letters both in the single-task and alternative dual-task condition, and not to use visual memory strategies during articulatory suppression. Instead, they were told to keep the phonologically recoded information in mind, although without any rehearsal of it. Finally, letter case was systematically changed between the targets and the probe in the memory conditions (see Fig. 1 for an example) in order to preclude a pure visual-matching strategy.
Using the SPM96 software package (http://www.fil.ion.ucl.ac.uk/spm/), the functional images acquired were realigned, corrected for motion artifacts, global signal intensity variation and low frequency fluctuations, normalized into the standard stereotactic space (MNI template) and spatially smoothed with a 12 mm/9 mm (for group/single subject analyses, respectively) full-width-half-maximum Gaussian kernel. For statistical analysis, the alternating task periods were modeled using a delayed box-car reference vector accounting for the delay of the BOLD (blood oxygen level dependent) response. Significantly activated pixels were searched for using the General Linear Model approach for time-series data. For this, a design matrix was defined for each subject comprising the following contrasts. The simple main effects of memory task performance were computed by contrasting MRI signal intensities during letter recognition with those during the letter case judgment task separately for each block condition. Interaction contrasts compared the simple main effect of memory task performance during articulatory suppression with the main effect(s) during single- and dual-task conditions (both separately and in combination), i.e. they determined specific effects of articulatory suppression on verbal memory processing, and were masked by the corresponding main effect(s). For group statistics, random effects analyses (Holmes and Friston, 1998) were performed on single subject contrast images using SPM99. This procedure accounts in particular for the inter-subject component of variance and, therefore, may be considered the currently most stringent and conservative statistical approach for fMRI data analysis permitting valid populational inference. Results are reported for brain activations that reached a significance level of P < 0.05, corrected for multiple comparisons either at the single-voxel level or based on the spatial extent of the activated region (Friston et al., 1994). Furthermore, the results of the group analysis were confirmed in individual analyses that allowed precise neuroanatomical identification of the brain structures activated in this study.
Both overt and silent articulatory suppression significantly reduced memory performance in the verbal item-recognition task (mean percentage of correct responses and standard deviations without articulatory suppression: 92.7 ± 6.3%; with overt articulatory suppression: 83.4 ± 8.2%; with silent articulatory suppression: 84.6 ± 11.4%; F = 7.71, P = 0.002). No significant differences were found between memory performance rates during overt and silent articulation (P = 0.68) or between reaction times in these conditions (overt articulation: 764 ± 104 ms; silent articulation: 734 ± 131 ms; P = 0.12).
As expected, the behavioral data that were ascertained during fMRI indicated again a significant reduction of memory performance during silent articulatory suppression, whereas alternating finger tapping had no significant effect (mean percentage of correct responses and standard deviations during single-task condition: 93.2 ± 4.5%; during silent articulatory suppression: 77.6 ± 9.1%; during alternating finger tapping: 91.1 ± 4.7%; F = 21.61, P < 0.001). No significant difference was found in reaction times (during single-task condition: 875 ± 175 ms; during silent articulatory suppression: 889 ± 146 ms; during alternating finger tapping: 853 ± 180 ms; F = 2.67, P = 0.09).
With regard to the functional imaging data, firstly the simple main effects of memory task performance were determined for each task variant. As predicted, the single-task condition revealed significant memory-related activations in brain regions known to subserve verbal rehearsal, i.e. in Broca's area (BA 44), the left lateral premotor cortex and the right cerebellum, as well as in the cortex along the left intraparietal sulcus (see Fig. 2A and Table 1B). The same areas were activated by memory requirements in the dual-task condition with alternating finger tapping (Fig. 2B). During articulatory suppression, by contrast, this pattern of brain activity changed greatly and activations related to memory task performance showed up in a different frontoparietal network including bilaterally the area along the anterior intermediate frontal sulcus and the inferior parietal lobule lining the banks of the ascending superior temporal sulcus (Fig. 2C and Table 1A). In addition, we found that in some other brain regions activity was enhanced by all working memory tasks independent of whether they were performed with or without articulatory suppression (Table 1C).
Subsequently, the specificity of the articulatory suppression effect on brain activity related to short-term memory requirements was statistically tested by means of interaction contrasts (see Materials and Methods) and proved to be highly significant. On the other hand, the reverse contrasts showed that the decrease of memory-related activity in brain areas underlying verbal rehearsal was also specific to the articulatory suppression condition (see the two right-most columns in Table 1). These specific effects of articulatory suppression are illustrated in Fig. 2D, which gives an overview of the brain activity patterns that were related to memory processes either under articulatory suppression (indicated in red), under non-interfering conditions (depicted in green), or in all of these task conditions (indicated in brown).
The results of the group analysis were confirmed by single subject analyses that, in addition, allowed exact neuroanatomical identification of the activated brain structures (see Table 1). For instance, in some of the individuals the anterior prefrontal activation observed during memory performance under articulatory suppression was found to be located along the anterior portion of the intermediate frontal sulcus (see blue arrows in Fig. 3B). In subjects in which this part of the sulcus is lacking, the same activation occurred in a corresponding anterior region at the top of the middle frontal gyrus. Overall, the activations related to memory performance during concurrent articulations were present in all subjects independent from the individual error rate or the relative memory performance reduction under articulatory suppression. Figure 3A–C shows these memory-related activations in three individual subjects differing gradually with regard to both the absolute error rates in the articulatory suppression condition (A: 39%; B: 22%; C: 11%) and the relative changes of memory performance due to articulatory suppression (as compared to memory performance in the single-task condition; A: 35% reduction; B: 22% reduction; C: 2% increase). Most importantly, the latter case (subject C) nicely illustrates that an increase of error rate is not a prerequisite for the occurrence of these activations. Furthermore, although subject B occupies an intermediate position with respect to the behavioral indices, it developed the strongest memory-related activations of all three subjects. Finally, we also calculated correlation coefficients between the error rates and reaction times in the articulatory suppression condition as well as the relative changes of these behavioral indices induced by articulatory suppression, on the one hand, and the effect sizes in the left and right anterior intermediate frontal sulci and inferior parietal lobules in the corresponding imaging contrasts (i.e. Mact(AS) and Mact(AS) – Mact(ST/DT); see Table 1), on the other, for the total group of 11 subjects. Significant correlations were only found between the effect sizes in the left and right anterior intermediate frontal sulcus (P = 0.019) and between the effect sizes in the left and right inferior parietal lobule (P = 0.013) which underlines the bilaterality of these brain activations. By contrast, there were no significant correlations between the effect sizes in these brain regions and the individual performance data.
This is the first neuroimaging study that investigated the neural correlates of verbal working memory both during single-task conditions and during articulatory suppression. In a behavioral experiment, it was shown that overt and silent articulatory suppression are equally effective in reducing memory performance in a verbal item-recognition task. In accordance with numerous previous studies that explored the functional neuroanatomy of verbal working memory (Paulesu et al.1993; Petrides et al.1993; Andreasen et al.1995; Awh et al.1996; Fiez et al.1996; Schumacher et al.1996; Cohen et al.1997; Jonides et al.1998), this verbal item-recognition task, when performed under single-task conditions, activated Broca's area, the left lateral premotor cortex, the right cerebellum as well as the cortex along the left intraparietal sulcus (Fig. 2A and Table 1B). The same areas were activated in a non-interfering dual-task condition with alternating finger tapping (Fig. 2B) indicating that dual-task performance per se had no relevant influence on the brain processes that subserve verbal short-term memory performance. In striking contrast to this, memory task performance under silent articulatory suppression was associated with activations in a different, prefronto-parietal network, and no significant memory-related activation was found in the above-mentioned areas activated under conditions without interference (Fig. 2C and Table 1A,B). It is important to note that all these activations were determined by subtracting from the respective memory condition the corresponding control condition, the only difference between these two tasks being the short-term memory requirements. Furthermore, the change of the memory-related activation pattern observed under articulatory suppression basically resulted from modifying the delay phase of the task whereas the external conditions during the presentation of targets and probes remained identical for each memory task. Thus, on one hand these data are consistent with the notion that articulatory suppression caused a specific competition for functional resources within the phonological loop that normally underlie verbal memory performance. This competition was reflected both by an impairment of verbal memory and by a lack of significant memory-related activity in brain areas that are known to subserve verbal rehearsal, e.g. Broca's area and the lateral premotor cortex. The drastic reduction of memory-related activation in these areas may thus be regarded as the neural counterpart of domain-specific interference within verbal working memory that has been observed in many previous behavioral studies (Shah and Miyake, 1996). On the other hand, however, it is even more remarkable that this interference of silent articulations with verbal rehearsal, in addition, led to a recruitment of other brain mechanisms located in different, especially anterior prefrontal and inferior parietal, brain areas (Figs 2D and 3). Several possible explanations have to be taken into consideration to account for this observation.
Possible Functions of the Prefronto-parietal Network Activated by Memory Task Performance under Articulatory Suppression
Since these anterior prefrontal and inferior parietal activations occurred specifically under the condition of articulatory suppression, it is reasonable to assume that they reflect brain processes recruited to deal with domain-specific interference in verbal working memory. Our behavioral data ascertained that this interference produced a significant increase of error rates, i.e. it reduced memory performance. It is important to note that this more specific interpretation of this behavioral effect is absolutely compatible with an alternative view that may ascribe the increased error rates in a simpler way to higher task difficulty. Task difficulty as indexed by performance data may be enhanced for various different reasons (Postle et al., 2000). In the present experiment, however, the critical factor that enhanced task difficulty was the introduction of domain-specific interference, while all other possible factors such as perceptual difficulty, memory load or the similarity between targets and non-targets were kept constant. Likewise, it is noteworthy that the prefronto-parietal network was even activated in individual subjects without an increase in error rates (see Fig. 3C), that the individual effect sizes in these anterior prefrontal and inferior parietal brain regions did not correlate with the individual performance data, and that a highly comparable visual working memory task for letter forms produced significantly less activation of these brain areas although being more difficult according to error rates than the phonological memory task variant which was also used in the present experiment [mean accuracy/reaction times in the phonological memory task: 76.1%/1032 ms; in the visual memory task for letter forms: 71.7%/1102 ms; see Gruber and von Cramon (Gruber and von Cramon, 2001)]. These data give clear evidence that activation of this network of prefrontal and parietal brain regions may not be sufficiently accounted for as being simply an effect of task difficulty, but that it is rather the memory demands under conditions of domain-specific interference that evoke this activation.
So, if domain-specific interference with the rehearsal mechanism of verbal working memory is the crucial factor eliciting activation of the prefronto-parietal network, the question arises as to how the brain deals with this interference, i.e. what the actual nature of the processes evoked by this specific task is. In other words, the question is: how does the brain manage to maintain verbal short-term memory performance in the presence of concurrent articulations that prohibit verbal rehearsal? Several alternative and, in part, complementary interpretations are possible. First, memory performance could have been maintained by switching between counting and rehearsal. In this case, the observed activations in the articulatory suppression condition may represent such switches between two processes. However, the present study was designed to completely prevent the subjects from any rehearsal by clear instructions to perform articulatory suppression at a paced and high articulation rate. In fact, both the introspective reports of the subjects and particularly the lack of significant memory-related activation in brain areas known to subserve verbal rehearsal (see Table 1B) indicate that the subjects were unable to rehearse the memorized letters when they performed the concurrent articulations according to the instructions. Therefore, it is not very compelling to assume switch processes between counting and rehearsal as the latter process obviously did not exist at all in this condition.
A second alternative is that executive processes were called up to coordinate the simultaneous performance of the two competing tasks. Given that some of the activations in question (e.g. in the anterior prefrontal cortex and in the anterior cingulate cortex) are similar to those associated with elements of executive control in previous studies (D'Esposito et al., 1995; Carter et al., 1998; Koechlin et al., 1999), this is a reasonable hypothesis. For example, in accordance with recent neuroimaging studies on the neural implementation of cognitive control (Carter et al., 1998; Botvinick et al., 1999; MacDonald et al., 2000) activation of the anterior cingulate cortex may represent processes subserving the detection of processing conflicts produced by domain-specific interference. However, as executive processes are usually considered to not have any storage functions (Baddeley, 2000), and since the ‘classical’ verbal working memory areas were not activated by memory performance during articulatory suppression (see Table 1B), it is likely that at least some of the other brain areas activated in this experimental condition, e.g. the anterior prefrontal cortex and the inferior parietal lobule, directly supported memory task performance during articulatory suppression by providing such memory resources. In this case, the question arises of whether these additional neuronal resources provided for memory performance are specific to the phonological domain or not. Correspondingly, one would have to consider whether there are other possible non-phonological strategies that the subjects might have used during articulatory suppression. First of all, it appears necessary to exclude that the memory-related activations during articulatory suppression may have been produced by visual working memory strategies. This, in fact, could not be definitely ruled out from the present study even though the subjects were explicitly instructed and, in addition, letter case was systematically varied to force them to respond on the basis of phonological identity and not visual form. Therefore, in order to differentiate the activations detected under articulatory suppression from visual working memory mechanisms, an additional fMRI experiment was conducted using similar tasks with silent articulatory suppression, during which colored letters in different fonts were presented and either the letters themselves or their colors or specific forms were to be remembered (Gruber and von Cramon, 2001). Replicating the results from the fMRI study reported here, the phonological task variant activated the same brain areas as depicted in Figure 2C. On a rough neuroanatomical level, visual working memory for colors or letter forms produced a similar fronto-parietal activation pattern. Nevertheless, there were statistically significant, domain-specific differences in the activations that were elicited by the different tasks. In particular, while the phonological task variant yielded strong activations along the anterior intermediate frontal sulcus and in the inferior parietal lobule, working memory for visual letter forms or colors preferentially activated more posterior prefrontal regions along the intermediate and superior frontal sulci as well as the superior parietal lobule. Thus, this study provided evidence that the activations that were also associated with memory under articulatory suppression in the present study did not result from visual working memory strategies (Gruber and von Cramon, 2001). Furthermore, the very short duration of target presentation should have strongly limited the possibility to resort to semantic memory strategies or to encode the items more deeply into intermediate or long-term memory. On the other hand, it cannot be excluded from this study that the activations under articulatory suppression may represent the recruitment of more ‘central’ memory resources that are not specific to the phonological domain, but which can be generally used when secondary tasks interfere with predominant verbal or visual memory strategies in a domain-specific way.
A Prefronto-parietal Network for Phonological Storage?
Further investigations are certainly needed to specify the functional contributions of the different brain regions involved in memory performance under articulatory suppression. There already exists, however, a large corpus of indirect evidence providing support for the assumption that the anterior prefrontal and inferior parietal brain activations under debate are related to phonological working memory, and not more generally to interference in working memory. First, several recent brain imaging studies in humans clearly failed to confirm this alternative assumption that dual-task interference may evoke specific executive processes located in distinct brain regions (Klingberg, 1998; Adcock et al., 2000; Bunge et al., 2000). Secondly, activations of the anterior prefrontal cortex similar to those in the present study have been repeatedly observed in a variety of verbal working memory experiments even if they did not require any coordination of concurrent tasks [see MacLeod et al. for a review (MacLeod et al., 1998)]. Most remarkably, not only this anterior prefrontal focus, but actually the complete pattern of memory-related brain activation found in the present study under articulatory suppression (Table 1A) appears identical to the activations that showed up in a very recent study when it was explicitly tested for phonological storage (Henson et al., 2000).
Beside this evidence from brain imaging studies in human subjects, strong support for a domain-specific memory function of the anterior prefrontal region in question is also provided by neurobiological investigations in non-human primates. On one hand, recent cytoarchitectonical data point to the intermediate frontal sulcus as the possible human homologue of the principal sulcus in macaques (Petrides and Pandya, 1999). On the other hand, this monkey brain area has been repeatedly demonstrated in lesion and neurophysiological studies to be involved in the maintenance of stimulus representations across delay periods during which the stimulus is absent (Goldman-Rakic, 1996). More specifically, recent investigations using microelectrode recording and anatomical tract tracing have led to the suggestion that the rostral principal sulcus, which appears to correspond to the anterior prefrontal area activated in the present study, may be involved in the mnemonic processing of non-spatial auditory and possibly also of species-specific vocal information (Hackett et al., 1999; Romanski et al., 1999). Further studies are needed to investigate whether or not cortical areas activated by phonological working memory can be anatomically differentiated from other, possibly adjacent human brain regions subserving more basic non-spatial auditory working memory processes.
Previous lesion studies have established a pivotal role of the left inferior parietal lobule for phonological storage [see Vallar and Papagno's review (Vallar and Papagno, 1995)]. Surprisingly, however, the inferior parietal lobule (among other brain regions presumably constituting a prefronto-parietal network) was only activated under articulatory suppression, and not when the subjects performed the memory task by intensive rehearsal (Table 1A), although the phonological store, as the presumed second component of the phonological loop, could usually be expected to be activated under these task conditions as well. [Such activation common to verbal working memory performance both with and without articulatory suppression was found in some other, frontal brain regions (see Table 1C). However, it appears very unlikely that any of these brain regions could be a candidate for the phonological store since so far there seems to be no empirical evidence that could support this claim. Instead, these frontal areas may rather be involved more generally in the implementation of a mnemonic task set or, as a common component of the two different brain circuits revealed in this study, they could mediate functional interactions between these circuits.] On the other hand, another parietal activation was found in only the rehearsal conditions along the left intraparietal sulcus (Fig. 2A,B). With respect to Talairach coordinates, this latter activation is very similar to many of those found in previous neuroimaging studies of verbal working memory [cf. table 1 of Becker et al.'s recent overview article (Becker et al., 1999)]. Overall, these ‘parietal findings’ can be interpreted in two different ways. First, in line with these previous reports, one could argue that the intraparietal area may represent the phonological store. Then, however, neither the rehearsal mechanism nor the phonological store would have been activated by verbal memory task performance under articulatory suppression in the present study (see Table 1B). Consequently, one would have to postulate that the prefronto-parietal network, which obviously had to support memory performance under articulatory suppression in some way, provided an additional short-term memory resource beside rehearsal and phonological storage, and also different from visual working memory mechanisms [as Gruber and von Cramon have demonstrated (Gruber and von Cramon, 2001)]. Most recently, Baddeley has offered a possible, though hypothetical solution for this problem by adding a new, multimodal storage component, the ‘episodic buffer’, into his working memory model (Baddeley, 2000). Accordingly, one may interpret at least some of the prefrontal and parietal activations observed under articulatory suppression in the present study as neural representations of this ‘episodic buffer’. On the other hand, if one considered alternatively [and in better agreement with the neuropsychological evidence; see Vallar and Papagno (Vallar and Papagno, 1995)] the dorsal supramarginal and angular gyri to be the crucial parietal regions for the function of the phonological store, then our findings may be taken as an indication that, independent from the activation state of phonological storage mechanisms, rehearsal alone may suffice to keep a certain amount of verbal information in mind. In this case, one may regard the other, intraparietal activation to be associated with rehearsal itself, a notion that seems reconcilable with previous neuroimaging studies of verbal working memory since these studies did not control for the exact rate of rehearsal or articulatory processes. This interpretation, however, is not consistent with the usual assumption that the phonological store is continuously refreshed by verbal rehearsal and, therefore, should have been activated in the present study in the single-task condition as well. Hence, a careful reconsideration of the empirical data that led to this assumption is needed in order to find out whether the same results could also be explained by a different kind of functional interaction between the mechanisms of verbal rehearsal and phonological storage. The double dissociation observed in the present study between articulatory and non-articulatory mechanisms of verbal working memory seems to suggest that the brain regions responsible for phonological storage, rather than being continuously ‘refreshed’ and co-activated, may receive input from the brain circuit involved in verbal rehearsal only in moments and situations in which subjects do not or even cannot rehearse any longer and in which, consequently, verbal memory performance has to rely on another mechanism. This view could also account for the occasional co-activation of the anterior prefrontal cortex and/or the inferior parietal lobule during verbal working memory tasks under single-task conditions [e.g. see MacLeod et al. (MacLeod et al., 1998)], because any underdetermination of the memory strategy to be used, be it due to the lack of clear instructions or the lack of compliance, renders a mixture of articulatory (rehearsal) and non-articulatory (storage) strategies very likely.
In conclusion, the present fMRI study has shown that verbal working memory task performance during articulatory suppression does not essentially rely on the ‘classical’ brain areas of the articulatory loop, but on a network of anterior prefrontal and inferior parietal brain regions that presumably underlie an alternative, non-articulatory mechanism for maintaining phonological representations, i.e. phonological storage. In the wider context of convergent evidence coming from neuroimaging studies in human subjects and neurobiological investigations in non-human primates, these findings may have interesting theoretical implications with respect to both the functional– neuroanatomical organization and the evolutionary development of human working memory (Gruber, 2001; Gruber and von Cramon, 2001).
I thank Prof. D. Yves von Cramon, Prof. Patricia S. Goldman-Rakic and anonymous reviewers for helpful comments on earlier versions of the manuscript.