Content uploaded by Naomi Langerock
Author content
All content in this area was uploaded by Naomi Langerock on Aug 09, 2021
Content may be subject to copyright.
1
The Maintenance of Cross-domain Associations in the Episodic Buffer
Naomi Langerock
1
Evie Vergauwe
2
Pierre Barrouillet
1
(1)University of Geneva, Switzerland
(2) University of Missouri, Columbia, USA
Address correspondence to:
Naomi Langerock
Université de Genève
Faculté de Psychologie et de Sciences de l’Education
40, Boulevard du pont d’Arve
1205 Genève Switzerland
Tel : 0041 22 379 92 41
Fax : 004122 379 92 29
Naomi.Langerock@unige.ch
DRAFT APPROVED FOR PUBLICATION IN JOURNAL OF EXPERIMENTAL PSYCHOLOGY:
LEARNING; MEMORY AND COGNITION.
Langerock, N., Vergauwe, E., & Barrouillet, P. (2014). The maintenance of
cross-domain associations in the episodic buffer. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 40(4), 1096.
https://psycnet.apa.org/record/2014-05765-001
2
Abstract
The episodic buffer has been described as a structure of working memory capable of
maintaining multi-modal information in an integrated format. Although the role of the
episodic buffer in binding features into objects has received considerable attention, several of
its characteristics have remained rather underexplored. This is the case for its maintenance
capacity limits and its separability from domain-specific maintenance buffers. The present
study addressed these questions making use of a complex span paradigm in which participants
were asked to maintain cross-domain (i.e. verbal-spatial) associations. The first experiment
showed that the capacity limit for these cross-domain associations proved to be lower than the
capacity limit for single features, and did not exceed three. Cross-domain associations and
single features depended however to the same extent on attentional resources for their
maintenance. The second experiment showed that domain-specific (verbal or spatial)
resources were not involved in the maintenance of cross-domain information, revealing a
clear distinction between the episodic buffer and the domain-specific buffers. Overall, in line
with the episodic buffer hypothesis, these findings support the existence of a central system of
limited capacity for the maintenance of cross-domain information.
Keywords: working memory, episodic buffer, cross-domain
3
Introduction
When testing Atkinson and Shiffrin’s (1968) assumption that the short-term store of the
modal model was a working memory (WM), Baddeley and Hitch (1974) concluded that the
idea of a WM comprising a single unitary store might be abandoned. Instead, they proposed
the well-known three-component model that distinguished between attentional control and
temporary storage, which was served by two systems, a verbal and a visuo-spatial one. This
model proved remarkably heuristic and, even though it underwent several evolutions
concerned with the specific relationship between its constituent subsystems, it remained
basically unchanged in its structure until the introduction of a new component, the episodic
buffer (Baddeley, 2000). This episodic buffer was intended to overcome the problems
encountered by the multi-component model. One of the main problems of the three-
component model was concerned with the fact that the representations constructed and
maintained by WM are essentially multi-modal, providing us with integrated and coherent
scenes of our environment. That is, the representations that constitute the content of our
conscious awareness are multi-modal in nature, as opposed to purely verbal or visuo-spatial.
In the three-component model, there was no subsystem that could provide temporary storage
for such kind of information. Of course, the central executive was conceived in the original
three-component model as playing a crucial role in this integrative function, but this system
was itself devoid of any function of storage. And thus, a new component was introduced, the
episodic buffer, which was conceived as comprising “a limited capacity system that provides
temporary storage of information held in a multi-modal code, which is capable of binding
information from the subsidiary systems, and from long-term memory, into a unitary episodic
representation” (Baddeley, 2000, p. 417).
More than ten years after its introduction, a survey of the literature reveals that among
the functions that were initially attributed to the episodic buffer, research has mainly
4
concentrated on binding, and this enquiry has, with rare exceptions, focused on binding within
a given domain of WM, either visuo-spatial (e.g., colors and shapes bound into objects; e.g.,
Allen, Baddeley, & Hitch, 2006) or verbal (e.g., words bound into chunks or sentences; e.g.,
Baddeley, Hitch, & Allen, 2009).This is in sharp contrast with one of the distinctive features
of this episodic buffer, its function of multi-modal information storage. Its capacity to
maintain multi-modal information as well as its limited maintenance capacity as opposed to
its binding capacity have remained largely neglected. What is the amount of multi-modal
information that can be maintained in face of distraction and interference? What are the
factors that affect the maintenance of this type of information? To what extent can the
episodic buffer be separated from the other storage systems of WM? These are the main
questions addressed in the present study.
The episodic buffer and the binding process
The need for a new component of WM came from a series of problems encountered by
the multi-component model, essentially from neuropsychological and developmental studies
(Baddeley, 2000). Though the hypothesis of a phonological loop accounted for a series of
phenomena, it appeared that patients with an auditory span of one digit were able to recall up
to four visually presented digits, suggesting some back-up store capable of integrating visual
and phonological information. In the same way, the preserved immediate prose recall in
densely amnesic patients suggested a temporary activation of LTM knowledge, but more
importantly the capacity to create mental representations within some back-up store
(Baddeley, 2000). Evidence in children for a capacity of reactivating memory traces before
their acquisition of adult subvocal rehearsal strategy along with the difficulty to give a good
account of maintenance mechanism in the visuo-spatial sketchpad pointed towards the
existence of some sort of general rehearsal that could involve the sequential attention to the
5
items to be recalled (Baddeley, 2000). Overall, the need appeared for a store capable of
maintaining information being integrated from both the two slave systems and from LTM.
According to Baddeley, Allen, and Hitch (2010), the main question concerning the
episodic buffer is its capacity to bind multi-modal information into unitary objects, concepts
or episodes. Interestingly, in Baddeley’s (2000) seminal article, it was acknowledged that
initially the central executive was given a crucial role in binding, ignoring the fact that the
central executive did not serve a storage function. The locus of the binding process concerned
a very uncertain issue, as in Baddeley, et al. (2010) it was acknowledged that the episodic
buffer was initially envisaged as a buffer with a substantial but unspecified degree of
processing capacity. At the introduction of the episodic buffer, it was conceived as a new
component of WM as well as a fractionation of the central executive. More precisely,
Baddeley (2000, p. 422) assumed that the episodic buffer could “provide the storage, and the
central executive the underlying processing for episodic memory”. Indeed, in the revised
model (Baddeley, 2000; Figure 1), it was proposed that information could not be fed into the
episodic buffer directly from the slave systems, but either from episodic LTM or through the
central executive, suggesting that the central executive was responsible for the binding
process. It is probably this uncertainty about the locus of the process of binding that led to
concentrate the research effort on this process and to leave the storage function of the episodic
buffer rather under investigated.
Assuming that everything within WM accesses the episodic buffer via the central
executive (see Figure 1), Baddeley and his colleagues developed a research program to test
the hypothesis that blocking the central executive would specifically disrupt binding. Two
domains were investigated: the binding of visuo-spatial features into objects and of words in
the comprehension and retention of prose. In the visuo-spatial domain, Allen, Baddeley, and
Hitch (2006) assessed through probe recognition the retention of either isolated features such
6
as shapes or colors, or combinations of these features into integrated objects. In the former
condition, participants were asked whether the probed color or shape was present in an
original array of four shapes or four colors (i.e., individual feature conditions), whereas in the
latter, they judged whether the probed feature combination had been present in an original set
of four colored shapes, requiring a binding of the constituent features (i.e., binding condition).
The involvement of the central executive was investigated by requiring participants to count
backward in threes during the encoding phase. Adding such a secondary task led to a
substantial and significant decrease in recall performance, but importantly this decrease was
no larger in the binding than in the individual feature conditions. The authors concluded that
binding features into objects does not involve more central executive resources than encoding
single features. Likewise, the binding of temporally or spatially separated features was not
more dependent on central executive resources than the encoding of already bound features
(Karlsen, Allen, Baddeley, & Hitch, 2010), not even when they were presented in different
modalities (e.g., visually and auditorily; Allen, Hitch, & Baddeley, 2009). These findings
were corroborated by a series of studies showing that the maintenance of feature bindings in
visual short-term memory does not require attention over and above that required for
maintaining individual features (Delvenne, Cleermans, & Laloyaux, 2010; Gajewski &
Brockmole, 2006; Johnson, Hollingworth, & Luck, 2008; Yeh, Yang, & Chiu, 2005).
Baddeley, Allen, and Hitch (2011) concluded from a review of these studies that the
perceptual system binds all features automatically and that binding does not appear to depend
on the central executive. Yet, not all studies on visuo-spatial binding would agree with this
conclusion, as non-automacity of binding visuo-spatial features has been observed in a
number of studies (e.g., Olson & Jiang, 2002; Postma & De Haan, 1996; Wheeler &
Treisman, 2002).
7
The other domain investigated involved binding in verbal WM and led to basically the
same conclusions. Brener (1940) observed that immediate word recall performance
dramatically increases when these words form part of a sentence compared with unrelated
words, reflecting the recall of larger chunks in the case of prose (Tulving & Patkau, 1962).
Using a variety of secondary tasks, Baddeley, Hitch, and Allen (2009) observed that
concurrent tasks had the same detrimental effect on recall of unrelated words (i.e., individual
feature condition) and sentences (i.e., binding condition), suggesting no major role of WM in
binding verbal information.
Overall, Baddeley et al. (2010) concluded from their studies that the episodic buffer
operates as a multidimensional but essentially passive store that is not responsible for the
formation of bindings, which appears to operate outside WM. Though this research program
constitutes an extensive investigation of the binding process on two different domains,
important questions remain. Initially, as we noted above, the episodic buffer was proposed to
tackle the problems encountered with the three-component model, the first of them being “the
way in which the various components of WM, each using a different code, could be
integrated” (Baddeley et al., 2010, p. 229). In this respect, the integration of visual features
into objects and of words into sentences are key questions of cognitive psychology, but it
could be argued that they do not concern the episodic buffer per se, because they do not
primarily require the integration of different codes and occur within the visuo-spatial and the
verbal systems respectively. Moreover, understanding how binding is achieved is an
important question, but the episodic buffer was also defined as a limited capacity system that
provides temporary storage of information. Nonetheless, the vast majority of studies focused
on very short retention intervals of 900 ms, while WM functioning obviously requires
maintaining information over longer periods of time. Few studies have investigated this
maintenance component of binding, and designs involving longer retention intervals led to
8
mixed results. Using intervals of 4100 ms, Morey and Bieler (2013) did not find any
interaction between attentional load and binding in the maintenance of color-shape
combinations, though Fougnie and Marois (2009), using even longer intervals of 6800 ms, did
observe that the recognition of bound visuo-spatial information was more disrupted by an
attentive tracking task than was the recognition of single features.
In summary, the research program intended to investigate the episodic buffer seems to
have neglected two crucial aspects of this new component of WM that are its storage function
and its role of integrating information from the slave systems of WM. In the next section, we
address the few studies that have investigated memory for cross-domain bindings through the
integration of verbal and visuo-spatial information.
Cross-domain binding
Several studies have provided evidence for an integrated instead of independent
maintenance of verbal and spatial information, thus suggesting the existence of the episodic
buffer. For example, Elsley and Parmentier (2009) as well as Prabhakaran, Narayanan, Zhao,
and Gabrieli (2000) presented participants with letters in locations for further recognition of
the letters and the locations. At test, participants were to indicate whether both the letter and
the location had been presented before, independently of their original association at study.
Both studies observed that exact letter-location combinations were faster and more accurately
recognized than recombined letter-location units. This is a result that cannot be accounted for
by a model that presumes a separate storage of verbal and spatial features. Further evidence
for an integrated maintenance of verbal and spatial features comes from studies showing that
effects usually observed with verbal memoranda affect memory for spatial information when
verbal and spatial information were shown integrated at study. For example, articulatory
suppression, which is known to have no effect on the maintenance of single spatial features,
has been shown to affect memory for spatial information when this information was presented
9
in an integrated format with verbal information and this binding explicitly had to be
maintained (Morey, 2009). The same phenomenon was observed with the phonological
similarity effect (Guérard, Tremblay, & Saint-Aubin, 2009). Thus, little doubt exists about the
interdependence of verbal and spatial maintenance.
However, and quite surprisingly, few studies have questioned the role of the central
executive in this cross-domain binding process. To our knowledge, the study by Elsley and
Parmentier (2009) is the sole study having investigated cross-domain binding while
manipulating attentional demands. It appeared that the already reported advantage at
recognition of exact letter-location combination compared with recombination disappeared
under concurrent acoustic memory load, suggesting a clear involvement of attention, and thus
of the central executive, in the binding of cross-domain information. However, this study only
tested implicit binding, and results could have been different if participants had been
explicitly asked to maintain letter-location associations. Indeed, it was shown by Morey
(2011) that memory performance differs quite a lot for intentional and incidental binding,
with intentional binding leading to increased feature recall. The underlying systems for both
types of binding might thus also be different.
The present study
Besides a lack of studies investigating the involvement of attentional resources in the
process of cross-domain binding, the maintenance of cross-domain information over longer
retention intervals has completely been neglected. Repovs and Baddeley (2006) assumed that
in order to further explore the episodic buffer and its role in cognition in the same way as the
other components of WM, two classes of tasks needed to be developed, namely measures of
capacity and interference tasks. However, to the best of our knowledge, there has been no
study focusing on the capacity of the episodic buffer or systematically manipulating the
nature of interfering tasks. Thus, it can be concluded that in contrast with the other
10
components of WM, the episodic buffer remains largely unexplored. The present study was
conceived as an attempt towards a better comprehension of the episodic buffer and its
functioning. Its originality concerns (1) its focus on the maintenance capacity of the episodic
buffer, in contrast to its binding capacity, and (2) the study of cross-domain associations, in
contrast to within-domain associations.
Our study had a twofold aim. First, we aimed at assessing the amount of cross-domain
information that can be maintained in the episodic buffer as well as the cognitive demand
incurred by this maintenance compared with the maintenance of single features. For this
purpose, in a first experiment, we used a complex span task in which participants had to
maintain verbal (i.e., letters), spatial (locations) or cross-domain information (i.e., letters in
location) while performing an attention-demanding task. Varying the cognitive load of the
secondary task allowed us to evaluate to what extent the maintenance of bound information is
more attention demanding than the maintenance of verbal or spatial features. The span
procedure informed us about the capacity of WM when maintaining cross-domain
information. Second, we aimed at investigating the separation of the episodic buffer from the
slave systems. For this purpose, a second study used a selective interference paradigm
involving the maintenance of cross-domain information while performing an attention-
demanding secondary task intended to produce additional verbal or spatial interference, or to
remain neutral in this respect. The hypothesis of a clear separation between the episodic
buffer and the slave systems as described in the multi-component model (Repovs & Baddeley,
2006) would predict maintenance of cross-domain information to be relatively immune to
selective interference compared with the maintenance of domain-specific information.
11
Experiment 1
The aim of this first experiment was twofold. Its first objective was to assess the
capacity of the episodic buffer by comparing the maintenance of cross-domain information
(i.e., binding verbal and spatial features) with the maintenance of individual features.
Participants were presented with a complex span task in which they had to maintain series of
letters, spatial locations, or letters in location. Previous studies focusing on the bound storage
capacity in the verbal and visuo-spatial domains have provided convergent results pointing
towards a limit of about three or four chunks. This was the case when studying the binding of
visual features into objects (Luck & Vogel, 1997; Vogel, Woodman, & Luck, 2001; but
failure to replicate by e.g., Wheeler & Treisman, 2002; Olson & Jiang, 2002) as well as when
investigating the capacity of storing verbal chunks comprising several words (Chen & Cowan,
2009; Cowan, Rouder, Blume, & Saults, 2012). Thus, the same limitation of about four can be
expected for the storage of objects integrating verbal and spatial features. However, an open
question remains concerning the comparison between visuo-spatial and cross-domain storage:
will the capacity of storing letters in location equate that of storing spatial locations alone? It
has been mentioned before that it is difficult to give a good account of rehearsal within the
visuo-spatial domain (Baddeley, 2000; 2012). This fact, in addition to some experimental
results, has led to the idea that the maintenance of visuo-spatial information might not be
subserved by some domain-specific slave system but by a domain-general mechanism that
can be assimilated with the episodic buffer. For example, when studying verbal and visuo-
spatial WM, Vergauwe, Barrouillet and Camos (2010) observed that verbal storage was more
affected by a verbal than a visuo-spatial interfering task, whereas visuo-spatial maintenance
was affected in the same extent by verbal and visuo-spatial interference. We interpreted these
results as suggesting the existence of some domain-specific system of maintenance for verbal
information, for example a phonological loop, whereas visuo-spatial WM would be devoid of
12
specific maintenance mechanism, relying on a domain-general attentional system (see Morey
& Mall, 2012, for similar conclusions). Now, and contrary to verbal spans, WM spans for
visuo-spatial information are usually rather low and do not seem to exceed four items
(Vergauwe, Barrouillet, & Camos, 2009, 2010). At the moment, it remains however
undecided whether WM spans for cross-domain information are lower than spans for visuo-
spatial information or not. On the one hand, it could be imagined that the episodic buffer can
maintain as many verbal-spatial chunks as it can hold spatial features, in the same way as
Luck and Vogel (1997) observed that storage capacity was limited to about four objects
regardless of the number of features each object comprises. On the other hand, it is also
possible that less verbal-spatial chunks can be maintained than single features. Cowan et al.
(2012) observed in the verbal domain that the number of chunks held in WM decreases as
their size increases, people recalling less chunks when made of triplets than of single words.
The number of elements in each chunk could thus matter. Oberaurer and Eichenberger (2013)
similarly observed that within visual working memory recognition accuracy decreased as the
number of features making up an object increased. The present experiment addressed this
question by comparing the maintenance of verbal-spatial associations with the maintenance of
their isolated components.
Second, we tested whether the maintenance of bound information requires attention
above and beyond the attention needed to maintain isolated components. Szmalec,
Vandierendonck and Kemps (2005) have shown that a pitch discrimination task requires
attentional resources while domain-specific verbal or spatial resources are not presumed to be
involved. Because of its domain-neutrality, this kind of task has been used in several other
studies (e.g., Elsley & Parmentier, 2009; Imbo, Vandierendonck, & Vergauwe, 2007) and was
also adopted in the present study. After each memory item (i.e., letter, spatial location, or
letter in location), participants were asked to judge the pitch, either high or low, of a series of
13
tones. The pace at which tones were presented was varied to create three levels of cognitive
load (high, medium, or low) with faster pace resulting in higher cognitive load. Barrouillet,
Portrat and Camos (2011) demonstrated that any task that requires attention, as is the case for
the pitch discrimination task, can be tuned to create different levels of cognitive load. The
hypothesis of the maintenance of cross-domain information within an episodic buffer fueled
by the attentional capacity of the central executive (Repovs & Baddeley, 2006) predicts lower
spans with increasing cognitive load of the secondary task. Such an effect has already been
observed on the maintenance of verbal and visuo-spatial information (Vergauwe et al., 2010;
Vergauwe, Dewaele, Langerock, & Barrouillet, 2012). However, the question of a greater
effect of the secondary task on memory for bound information due to an additional cost
associated with the maintenance of the binding itself remains open. A specific cost of this
maintenance would result in an interaction between the cognitive load of the secondary task
and the type of memoranda, with a stronger effect on the maintenance of bound information
compared with the maintenance of letters or spatial locations. Of course, we reviewed above
several studies that have shown that retaining bound objects in WM does not require attention
over and above that required for maintaining featural information. However, these studies
focused on visuo-spatial WM and often concerned bindings of features such as shape and
color. As suggested by Baddeley et al. (2011), these bindings could be too basic and
automatic to be affected by concurrent cognitive load. A recent study by Ecker, Maybery and
Zimmer (2013) has suggested a difference in the automacity of binding based on the
perceived coherence of the to-be-bound features. Features that are perceived as belonging to a
same object are defined as intrinsically related (e.g., color and shape) and resulted in an
obligatory and automatic binding and retrieval of the binding. Features that are extrinsically
related are part of the same global encoding, but are not perceived as inherent characteristics
of a same object (e.g., shape and location). This extrinsic relation between features did
14
however not result in this kind of obligatory and automatic binding and retrieval as was the
case for intrinsically related features. It remains thus possible that the maintenance of cross-
domain bindings between verbal and spatial elements involves an additional attentional
demand as location is typically defined as an extrinsic feature. Nevertheless, automatic
binding of location to letters (but not of letter to locations) has been observed in several
studies focusing on verbal-spatial associations (e.g., Campo, et al.,2010; Guérard, Morey,
Lagace & Tremblay, 2013). We adopted the methodology used in several studies, i.e. the
maintenance of either single or bound features was compared under different levels of
concurrent attentional demand (e.g., Allen et al., 2006; Morey & Bieler; 2013).
Method
Participants
Eigthy-five students (mean age = 22.06 years, SD = 5.63 years, 73 females) were given
course credits for participation. The experimental session lasted between 30 and 60 minutes.
Due to this rather lengthy procedure, each participant accomplished only one maintenance
domain condition. Maintenance domain (verbal, spatial, or cross-domain) was thus
manipulated between subjects while attentional demand (low, medium, or high cognitive
load) was manipulated within subjects.
Materials and procedure
The complex span tasks involved the presentation of series of two to seven letters,
spatial locations, or letters in location in ascending length. Letters were consonants (W, Y and
Z were excluded) presented in a square centered on screen. Spatial series consisted of squares
successively lighting up in grey and filled with the same letter X among sixteen possible
locations indicated by 16 empty squares randomly distributed on the screen to avoid verbal
coding of their position. Cross-domain maintenance items consisted of letters displayed
15
within these squares lighted up in grey. In each memory condition, series were
counterbalanced across cognitive load conditions in such a way that each series appeared
equally often for each cognitive load condition over all participants. The processing task was
a choice reaction task in which participants had to decide by pressing appropriate keys
whether a presented tone was low (262Hz) or high (524 Hz) in frequency (Szmalec, et al.,
2005). The occurrence of high and low tones was random.
By making use of a complex span task, the secondary task was only to be performed
during the maintenance stage and not during encoding or retrieval. Studies using a secondary
attention-demanding task during explicit binding and its subsequent maintenance (e.g.,
Karlsen, et al., 2012) and retrieval (e.g., Allen, et al., 2006) did not show any difference in
terms of attentional demand by encoding, maintenance or retrieval for bound items as
compared to single items. In order to measure the attentional demand for maintenance over
extended retention intervals, the secondary task was administered only during the
maintenance stage of the task.
The experiment began with training for the maintenance task, for the processing task
and for the combination of both. Figure 2 shows the trial design. Each trial began with a 750
ms indication of the pace of the processing task (“lent” (slow), “moyen” (medium), or
“rapide” (fast) displayed on the center of the screen) followed by a centrally displayed
asterisk for 750 ms. Hereafter, the first maintenance item was presented for 1500 ms
followed, after a blank screen of 500 ms, by a series of tones presented through headphones at
a computer-paced rate. Each tone was presented for 200 ms. The low, medium and high
cognitive load conditions were the same as in Vergauwe et al. (2010) and created by
presenting either 4 tones at a rate of one tone every 2000 ms, 4 tones at a rate of one tone
every 1293 ms, or 8 tones at a rate of one tone every 1000 ms respectively. Thus, cognitive
load was varied either by manipulating the number of items within a fixed interval of 8
16
seconds or by manipulating the rate at which a fixed number of items was presented. A new
maintenance item was presented immediately after the processing phase, followed by a 500
ms blank screen and a new processing phase. At the end of the trial, the word “RAPPEL”
(RECALL) appeared on screen and subjects were to recall the maintenance items in correct
order. For the series of letters, they typed the consonants in succession and validated their
responses by pressing “enter” after each letter. For spatial locations, they used the mouse to
click successively on each location. Each click on a square turned it grey until the “enter” key
was pressed to validate the response and recall the following location. For letters in locations,
the same procedure was used, but participants typed the appropriate letter after having clicked
the location.
Series of two to seven memory items were presented in ascending length with three
trials per cognitive load condition, resulting in 9 trials per length presented in random order.
For a given length, if participants correctly recalled two trials at a given cognitive load, the
third trial was omitted and counted as correct. If participants recalled no more than one trial
for each cognitive load condition correctly, the experiment terminated after this block. These
rules were implemented in order to avoid a drop in motivation due to too easy or too difficult
series. A span score was calculated adding 1/3 for each trial correctly recalled (all letters,
locations or letters in location in the correct order). Each participant started with a basis score
of one (as series of one item were not presented because of their estimated ease). The
maximum span score for each cognitive load condition was thus 7.
Results
Four participants were excluded from further analysis. One participant in the verbal
condition did not reach the predetermined criterion of 80% correct on the processing task and
two others had recall scores exceeding two standard deviations from the mean on at least one
17
of the cognitive load conditions. This was also the case for one participant in the spatial
condition. Each condition contained thus 27 participants for further analysis. A 3
(Maintenance Domain: verbal, spatial, cross-domain) x 3 (Cognitive Load: low, medium,
high) repeated measure Analysis of Variance (ANOVA) was performed with maintenance
domain as between-subject and cognitive load as within-subject factors. There was a
significant effect of maintenance domain, F (2, 78) = 59.32, p < .001, η
2
= .60 (Figure 3).
Planned comparisons showed that verbal recall was significantly better than spatial recall, F
(1, 78) = 44.74, p < .001, η
2
= .37 (mean spans of 5.76 and 3.92 respectively), which was in
turn significantly better than cross-domain recall (mean span of 2.79), F (1, 78) = 16.81, p <
.001, η
2
= .18. There was also a significant effect of cognitive load F (2, 156) = 16.12, p <
.001, η
2
= .17 (mean spans of 4.36, 4.21, and 3.91 for the low, medium, and high cognitive
load respectively), with a highly significant linear trend, F (1, 78) = 26.58, p < .001. This
linear trend accounted for 96 % of the variance associated with the effect of cognitive load.
There was no interaction between maintenance domain and cognitive load, F (4, 156) < 1. As
the repeated measure ANOVA did not allow us to reject the null-hypothesis on the absence of
an interaction, we calculated the posterior probability that the data favor the null-hypothesis.
To do so, we used the Bayesion information criteria (Masson, 2011) which resulted in p
BIC
(H
0
|D) = .99.
Discussion
According to Repovs and Baddeley (2006), maintenance of the information within the
episodic buffer would rely on the attentional resources of the central executive. Consequently,
the amount of information held in the episodic buffer would be affected by the amount of
attention captured by concurrent processing. In line with this assumption, we observed that
WM span for cross-domain information decreased when the cognitive load induced by
18
concurrent response selections increased, a phenomenon already observed with a variety of
memoranda (Barrouillet et al., 2011; Barrouillet & Camos, 2012). This suggests that the
resources needed for maintenance of cross-domain information were used by the processing
component of the complex span task. However, there was no interaction between cognitive
load and the nature of memoranda, either verbal, spatial, or cross-domain, indicating that the
maintenance of cross-domain information involves no additional demand over and above that
required for the maintenance of its constituents. Thus, as Baddeley et al. (2011) stated,
binding features into composite memory traces at encoding does not seem to involve
attentional resources, but in addition the present results show that the maintenance of these
bindings themselves during extended periods of time is not more costly. According to the
study by Ecker, et al. (2013), extrinsically related features (as is the case for letters in
locations) should not result in an obligatory and automatic retrieval of the binding.
Nevertheless, they showed that in an explicit binding task, the association between extrinsic
features did not come up automatically but could anyhow be retrieved on demand.
The present results allow us to specify the conclusions drawn by Baddeley et al. (2010)
from the exploration of the episodic buffer that he described as a passive store. The episodic
buffer is passive in that sense that it does not seem to play any active role in the binding
process itself, which is probably automatic and resulting from attentional focusing at
encoding (Cowan, 1995; 2005). Cowan has suggested that chunking and learning are
functions of the focus of attention, elements focused on at the same time being tied together
and to their context. Anticipating the concept of episodic buffer, Cowan (1995) suggested that
these new links would comprise an episodic record that becomes part of long-term memory.
The episodic buffer could also be considered as passive because, contrary for example to the
phonological store that is coupled with an articulatory loop, it is devoid of any specific
mechanism of maintenance, relying only on the executive resources of the central executive.
19
However, it could also be argued that the mechanism of attentional refreshing described by
several theories of WM (Barrouillet, Bernardin, & Camos, 2004; Cowan, 1999, 2005) and for
which neural correlates have been identified (Raye, Johnson, Mitchell, Greene, & Johnson,
2007) constitutes the mechanism by which the representations held in the episodic buffer are
maintained in an active state. This refreshing mechanism refers to the retrieval of memory
traces by attentional focusing. In this sense, the episodic buffer does not appear as a passive
store, or at least no more passive than the phonological loop.
The other main finding revealed by this first experiment concerns the capacity of the
episodic buffer. Concerning the maintenance of either verbal or spatial items, we observed
higher WM spans for verbal than for spatial information. This difference might have occurred
due to a difference in distinctiveness between verbal and spatial stimuli (Murdock, 1960),
with spatial locations being less discriminable. Another option is a difference in maintenance
mechanism between verbal and spatial stimuli. While there is ample evidence for the
existence of a phonological loop performing verbal rehearsal in order to maintain verbal
information, the nature of the rehearsal mechanism for visuo-spatial information is less clear
(Baddeley, 2012). Several authors even argue that the maintenance of visuo-spatial
information would be devoid of specific mechanism (Vergauwe et al., 2010; Vergauwe,
Camos, & Barrouillet, submitted; Morey & Mall, 2012). According to this assumption, the
maintenance of visuo-spatial information would rely on the sole attentional resources
assumed to fuel the episodic buffer, while the maintenance of verbal information could
benefit from these attentional resources as well as from verbal rehearsal (Camos, Lagner, &
Barrouillet, 2009). In line with previous estimates (Cowan, 2001, 2005), our participants were
able to maintain about 4 spatial locations. However, WM spans for objects integrating verbal
and spatial information did not exceed 3 objects. It should be noted that this estimate is
slightly lower than the one deducted from a study by Cowan, Saults, and Morey (2006). Their
20
task might though have been slightly easier as children and adult participants had to
reconstruct the association between three to seven names and three to seven spatial locations,
instead of recalling the associations. However, their result suggested similarly that this cross-
domain maintenance limit is lower than a spatial or name span. Where does this difference
come from?
Though the episodic buffer remains, as Baddeley et al. (2010, p 240) stated, a “shadowy
concept”, it is assumed to hold a limited number of chunks. This makes the episodic buffer
akin to the focus of attention described by Cowan (2001) or the object files hypothesized by
Kahneman, Treisman, and Gibbs (1992). Accordingly, it could be imagined that its capacity is
defined in terms of the number of objects that can be maintained rather than in terms of the
number of their constitutive features. Studies that have compared capacity for individual
features with capacity for objects have led to divergent results. In a recognition task of
memory items presented sequentially, Allen et al. (2006) reported that the retention of color-
shape bindings was inferior to the recognition of either color or shapes, even if the memory
for feature conjunctions was no more affected than the memory for features by a demanding
concurrent task. Our task involved recall instead of recognition, but we also observed poorer
memory for feature conjunctions than single features whereas the two types of memoranda
were affected in the same way by a concurrent attentional demand. However, and in line with
the conception of a capacity of the episodic buffer in terms of chunks or objects, Chen and
Cowan (2009) reported a constant capacity of three chunks made either of singletons or
previously learned pairs of words provided that covert verbal rehearsal is prevented. Luck and
Vogel (1997) made the same observation in the visuo-spatial domain, objects defined by a
conjunction of features being retained just as well as single-feature objects.
However, more recent investigations indicated that the size of the chunks to be
maintained could matter. Cowan et al. (2012) observed that the number of chunks maintained
21
in verbal WM decreases as their size increases. Whereas participants were able to maintain
slightly more than 3.5 chunks containing single words, this number dropped below 3.5 for
chunks made of pairs of words, and was about 2.5 for word triplets. Cowan et al. (2012)
invoked some chunk decomposition factor by which within-chunk associations fail or a given
chunk gives rise to several chunks that can only be remembered by occupying different slots
in WM. The same phenomenon could occur here. Although only about three chunks were
remembered, participants often correctly recalled an additional letter or location in isolation.
The spatial and verbal span scores of the cross-domain maintenance condition (i.e., the
maximum number of locations and letters correctly recalled whatever the recall accuracy of
their counterpart) were thus higher (4.02 and 3.09 for the mean verbal and spatial span) than
the cross-domain maintenance span (i.e., the number of chunks completely and correctly
recalled; mean span score 2.79), corroborating the idea of chunk decomposition. However,
these spatial and verbal span scores of the cross-domain maintenance where lower than the
spans observed in the pure spatial and verbal maintenance conditions. These findings seem at
odds with the hypothesis suggested by Cowan et al. (2006) of the maintenance of cross-
domain information as separated features held in parallel in their respective domain-specific
buffers (i.e., letters and spatial locations separately maintained) and simply associated at
recall on the basis of order information. Indeed, such a strategy would result in recalling as
many associations as items retained in the shortest ordered list, that is the list maintaining the
spatial locations. We have seen that this was not the case. Nonetheless, at least two
alternatives remain. First, it could be supposed that verbal and spatial features are maintained
in separated buffers, but that these buffers are fueled by a common resource. This would
result in a reduced number of items that can be maintained when verbal and spatial items have
to be held simultaneously (Morey & Cowan, 2004). Second, it could be supposed that verbal
and spatial features are bound in a buffer separated from the domain-specific systems, which
22
is the hypothesis of the episodic buffer by Repovs and Baddeley (2006). The present results
cannot disentangle these two possibilities. Therefore, the following experiment addressed
more directly the construct of the epsisodic buffer as a separate entitiy, dissociated from the
domain-specific maintenance buffers, through the selective interference paradigm.
Experiment 2
The selective interference paradigm has played a main role in establishing the
separability of the different slave systems hypothesized by the multi-component model
(Baddeley, 1986). Its underlying logic is that, if there exists different systems devoted to the
maintenance of either verbal or visuo-spatial information, then verbal activities would
selectively interfere with verbal maintenance while leaving visuo-spatial maintenance
relatively unaffected, whereas visuo-spatial activities would selectively interfere with visuo-
spatial maintenance but not (or less) with verbal maintenance. In the present experiment, we
extended this logic to the study of the episodic buffer. If cross-domain bindings are
maintained in some distinct buffer separated from both the verbal and the visuo-spatial
systems and fueled by a general-domain attentional resource, maintenance of bindings should
be affected by a concurrent attentional demand, but would remain immune from both verbal
and visuo-spatial interference. If the maintenance of cross-domain associations would anyhow
prove to be prone to verbal or spatial domain-specific interference, then this would question
the existence of the episodic buffer, or at least its separation from the verbal and visuo-spatial
maintenance systems.
Both Morey (2009) and Guérard, et al. (2009) showed spatial features to be subject to
domain-specific verbal interference when these were to be maintained integrated. While this
finding supports an integrated verbal-spatial maintenance, this does not exactly fit with the
hypothesis of an independent structure responsible for the maintenance of cross-domain
23
associations. However, Morey also showed verbal features to be less prone to verbal domain-
specific interference when information had to be maintained integrated. This in contrast does
seem to fit with independent structure responsible for the maintenance of cross-domain
associations. For the maintenance of cross-domain associations, domain-specific interference
seemed thus to act more between domains but less within domains in comparison with the
maintenance of isolated features.
In order to clarify the influence of domain-general and domain-specific interference on
the maintenance of cross-domain associations, this second experiment will investigate its
influence on the level of cross-domain associations, instead of on the feature level as was
done by Morey (2009) and Guérard, et al. (2013). We are above all interested in the
maintenance of cross-domain associations as a whole and not the maintenance of its isolated
components.
In this experiment, participants performed a WM span task in which they were asked to
remember letters in spatial locations while performing choice reaction time tasks intended to
produce verbal or spatial interference, or to remain neutral in this respect. In order to
appropriately differentiate between attentional and domain-specific interference, all these
distracting tasks required response selection and involved thus an attentional demand that was
manipulated to establish their detrimental effect on the maintenance of bindings. This
attentional demand was kept approximately constant across tasks by presenting the same
number of distractors at the same rates. However, these distractors were varied in nature to
selectively interfere with verbal or spatial maintenance, or to remain neutral in this respect,
resulting in a verbal, a spatial, and a neutral interference condition respectively. For this
purpose, participants were asked to perform either semantic judgments on words, spatial
judgments about the length of a line compared with the interval between two dots (both tasks
are derived from Vergauwe et al., 2010), or judgments about the pitch of tones as in
24
Experiment 1. The hypothesis of an episodic buffer separated from the slave systems and
fueled by attentional resources predicts that the maintenance of verbal-spatial bindings should
be affected by variations in the attentional demand of the distracting tasks but should remain
unaffected by the nature of the distractors to be processed and the specific interference they
elicit. Consequently, the neutral condition was predicted to be as disruptive as the verbal and
the spatial conditions. A control experiment established that the distracting tasks were
appropriate to produce the intended selective interference.
Method
Participants
Ninety-one (mean age = 21.48 years, SD = 3.53 years, 77 females) undergraduate
students were given course credits for participation. The experimental session lasted between
45 and 60 minutes. Each participant was randomly attributed to one of the three processing
domain conditions. Processing domain (neutral, verbal or spatial) was thus manipulated
between subjects while attentional demand (low–medium–high cognitive load) was
manipulated within subjects.
Materials and procedure
The three complex span tasks (incorporating a verbal, a spatial or a neutral processing
task respectively) had the same structure as the cross-domain maintenance condition used in
Experiment 1. Based on the performance from Experiment 1, series of letters in location
ranged from one to five. In each memory condition, series were counterbalanced across
cognitive load conditions in such a way that each series appeared equally often for each
cognitive load condition over all participants. The three processing tasks had the same
structure and the cognitive load was manipulated in the same way as in Experiment 1, with
either 4 distractors at a rate of one distractor every 2000 ms, 4 distractors at a rate of one
25
distractor every 1293 ms, or 8 distractors at a rate of one distractor every 1000 ms for the low,
medium, and high cognitive load conditions respectively. In the verbal task, distractors were
nouns selected out of 12 animal and 12 non-animal nouns auditorily presented in a random
order, with presentation times ranging from 440 to 660 ms depending on word length.
Participants decided by pressing keys whether the noun presented was an animal or not. The
spatial condition consisted in a spatial fit task involving 24 white boxes containing a black
horizontal line centrally displayed on screen and two black square dots positioned on the same
horizontal plane as each other, either above or below the horizontal line. The line varied in
length and the distance between the dots was chosen in such a way that, for half of the boxes,
the line could fit into the gap between the dots. These distractors were displayed on screen for
1333 ms, 862 and 667 ms and followed by blank screens of 667 ms, 431ms, and 333 ms in the
low, medium and high cognitive load conditions respectively. Participants were instructed to
decide whether or not the line could fit into the gap by pressing appropriate keys. The neutral
condition involved the same tone discrimination task as in Experiment 1.
In each task, the trials had the same structure as in Experiment 1 and the same
procedure was followed. The only difference concerned the presentation time of the
memoranda, which was reduced to 1000 ms instead of 1500 ms. Series of one to five memory
items were presented in ascending length with nine trials per length (three trials for each
cognitive load condition in random order) for a total of 45 trials. Each series correctly recalled
(all letters recalled in correct order in their exact location) was scored 1/3 for a maximum
span score of 5 for each cognitive load condition.
Results
One, eight and two participants in the neutral, verbal and spatial processing task
condition were excluded from further analysis as they did not reach the predetermined
26
criterion of 80% correct responses in the processing task. Another three, two and two
participants were excluded as their scores exceeded two standard deviations from the mean on
at least one cognitive load condition. This left us with 24, 24 and 25 participants in the three
conditions of the processing task respectively.
A 3 (Processing Domain: neutral, verbal, spatial) x 3 (Cognitive Load: low, medium,
high) repeated measure ANOVA was performed with processing domain as between subject
and cognitive load as within subject factor. There was a significant effect of cognitive load, F
(2, 140) = 55.38, p < .001, η
2
= .44 (Figure 4). Planned comparisons showed that recall in the
low cognitive load condition was significantly better than in the medium cognitive load
condition, F (1, 70) = 37.71, p = < .001, η
2
= .31, which was in turn significantly better than
in the high cognitive load condition, F (1, 70) = 24.60, p < .001, η
2
= .26. There was no
significant effect of processing domain, F (2, 70) < 1, and the interaction between processing
domain and cognitive load was not significant either, F (4, 140) = 1.36, p = .25
1
. For both
null-effects, we calculated the posterior probability favoring the null-hypothesis. This
probability was p
BIC
(H
0
|D) = .98 for the null-effect of processing domain and p
BIC
(H
0
|D) =
.99 for the interaction between processing domain and cognitive load. The difference that
occurred at a descriptive level between the three conditions of processing task at a high
cognitive load did not reach significance, F (2, 70) = 1.42, p = .25, even when the neutral
condition was contrasted with the verbal F (1, 70) = 1.20, p = .28
or spatial conditions, F (1,
70) = 2.75 , p = .10.
Discussion
Two important results came out of this experiment. First of all, the assumption of the
episodic buffer to be dependent on attention was once more confirmed. Varying the
attentional demand of the three intervening tasks had a detrimental effect on recall with higher
27
cognitive load resulting in poorer recall. Like the maintenance of individual features, the
maintenance of cross-domain information depends thus clearly on attention. On the contrary,
no evidence was shown for domain-specific resources to be involved in the maintenance of
cross-domain information. Neither the verbal nor the spatial processing task was able to create
any more interference than the neutral processing task. The three interfering tasks had the
same detrimental effect on the maintenance of cross-domain information with no indication of
selective interference.
This has important implications for the concept of the episodic buffer. The verbal
processing task was supposed to create verbal interference in the same way as the spatial
processing task was supposed to create spatial interference, that is over and above their
attentional demands, whereas the effect of the neutral processing task was supposed to be
restricted to its attentional demand. The observation that the maintenance of information in
the episodic buffer is not more prone to verbal or spatial than to neutral interference might
suggest the episodic buffer is indeed a storage system separated from the phonological loop
and the visuo-spatial sketchpad, for which selective interference has widely been shown
(Baddeley, 1986). One might, however, question the capacity of our tasks to create the
intended domain-specific interference. A control experiment was designed to discard this
possibility of non-adequacy of our tasks.
Control experiment
The objective of this control experiment was to show the adequacy of the processing
tasks used in experiment 2. As no difference was found between the three processing tasks, it
has to be established that this is not due to the processing tasks not being efficient in their
objective.
28
The verbal processing task is supposed to create verbal interference over and above its
attentional demands just like the spatial processing task is supposed to create spatial
interference over and above its attentional interference. While visuo-spatial interference has
been demonstrated on several occasions (e.g., Logie, 1986; Shah & Miyake, 1996), other
studies did not show any evidence for a visuo-spatial processing task to produce selective
interference (Vergauwe et al, 2010, Bayliss, Jarrold, Gunn & Baddeley, 2003). Different
patterns of selective interference between the verbal and visuo-spatial domain have
consequently questioned the existence of a domain-specific visuo-spatial resource (Morey,
2009; Morey & Mall, 2012). As this actual debate is not to be resolved here, we choose to
focus only on the adequacy of the verbal processing task to create verbal interference.
In Experiment 2, the verbal task had the same effect on the maintenance of cross-
domain information as the spatial task. To confirm the ability of the verbal task to create
domain-specific interference, it has to be demonstrated that it is more disruptive on the
maintenance of pure verbal information than a spatial task, thus revealing a selective
interference. This was the purpose of this control experiment. We compared the relative
impact of a verbal and a spatial processing task on the maintenance performance of single
verbal features on the one hand and on the maintenance of cross-domain information on the
other. This was done under a medium cognitive load. If it is indeed observed that the
maintenance of cross-domain associations is affected to the same extent by the verbal and
spatial processing task, while the maintenance of single verbal features is more affected by
the verbal than by the spatial processing task, then we could indeed conclude that our verbal
processing task is apt to create verbal interference.
29
Method
Participants
Forty-six (mean age = 21.93 years, SD = 6.55 years, forty-four females) undergraduate
students were given course credits for participation. Maintenance domain (cross-domain or
verbal) was manipulated between subjects while the nature of the processing task (verbal or
spatial) was manipulated within subjects.
Materials and procedure
Series of two to seven cross-domain (letter in location) items or verbal features were
presented in a complex span task. The presentation mode of the memoranda was the same as
in experiment 1. The processing task was either verbal or spatial in nature. The verbal
processing task was the semantic decision task used in Experiment 2. The spatial processing
task was the spatial fit task equally used in Experiment 2.
After training on the memory task, the processing task and a combination of both, the
experiment started. Participants were first presented with an indication of the processing task
this trial would incorporate. Then a fixation cross was displayed for 750 ms, followed by the
presentation of the first memorandum for 1000 ms and a 500 ms blank screen. Then the
processing phase started. During the processing phase, 4 items to process were presented in a
5172 ms interval, equalizing the medium cognitive load in the previous two experiments.
Hereafter a new memorandum was presented and followed by a processing phase. This
continued until the word RAPPEL (RECALL appeared on screen). At this moment
participants had to recall the memoranda in the same way as they did in experiment 1. Verbal
features were entered using solely the keyboard, cross-domain items were entered using the
mouse and the keyboard. In order to keep the performances on the processing task above
80%, participants received feedback on their processing task accuracy after each trial.
30
For each list length, three trials incorporating the verbal processing task and three
trials incorporating the spatial processing task were randomly performed. Scores were
calculated according to the span procedure. Each series correctly recalled gave rise to 1/3 of a
point. As the series of 1 memorandum were omitted, one point was added to the scores. The
maximum score by processing task was thus seven.
Results and Discussion
Two participants in the cross-domain maintenance condition and two participants in
the verbal maintenance condition were excluded as they did not reach the criterion of 80%
correct on the verbal or spatial processing task. Furthermore, one participant in the cross-
domain maintenance condition and one participant in the verbal maintenance condition were
excluded as their performances exceeded 2 standard deviations from the mean. A two
(Maintenance domain: cross-domain or verbal) by two (Processing task: verbal or spatial)
repeated measures ANOVA was performed with the nature of the processing task as within-
subject and the maintenance domain as between-subject factor. As expected, verbal
maintenance was significantly higher than cross-domain maintenance, F (1, 38) = 184.50, p <
.001, η
2
= .83 (Figure 5). The effect of the nature of the processing task was also significant, F
(1, 38) = 23.46, p < .001, η
2
= .38, as was the interaction, F (1, 38) = 17.53, p < .001, η
2
= .32.
A further analysis of this interaction showed that the verbal processing task had a more
detrimental effect than the spatial processing task in the verbal maintenance condition, F (1,
38) = 40.78, p < .001, η
2
= .52, but not in the cross-domain maintenance condition, F < 1.
The control experiment showed a clear selective interference, confirming the adequacy
of the verbal processing task to create verbal interference. Once again, the maintenance of
cross-domain associations was not prone to this verbal interference, endorsing the hypothesis
31
that the episodic buffer is an independent maintenance system, well separated from the
domain-specific maintenance buffers.
So notwithstanding the absence of effect of the domain of the processing task in
Experiment 2, the control experiment was able to valorize these results. As stated by
Baddeley (2012), negative results can be as valuable as positive results in guiding the
development of working memory models.
General discussion
The goal of the present study was to investigate the underexplored characteristics of
the episodic buffer, i.e., its maintenance capacity and its separability from domain-specific
maintenance systems. We chose to tackle these issues through the method of WM span
measure and selective interference. For this purpose, we used a complex span task paradigm
in which participants were asked to memorize verbal-spatial cross-domain associations (i.e.,
letters in location) while performing processing tasks varying in their verbal, spatial, and
attentional demands. In the first experiment, it was shown that WM spans are lower for
verbal-spatial associations than for their isolated verbal or spatial components. Irrespective of
these capacity limits, a decrease in attentional availability resulted in an equal drop in
maintenance performance for single features and cross-domain items, suggesting thus no
additional attentional demand to maintain cross-domain multi-feature information. Although
this first experiment left open the possibility that the verbal and spatial features composing the
cross-domain memoranda were separately maintained in domain-specific buffers instead of in
some episodic buffer, the second experiment discarded this option. A selective interference
paradigm showed the maintenance of cross-domain information to be solely dependent upon
attentional resources, with no domain-specific interference. This result reinforces the
32
hypothesis of independence of the episodic buffer from the domain-specific maintenance
buffers. In the following, we discuss these main findings.
The first issue addressed in this study concerned the capacity limit of the episodic
buffer conceived as a system for the assimilation of information from different domains into
integrated objects. Consequently, we focused on the maintenance of verbal-spatial
associations. The results of the first experiment clearly indicated the capacity limit of the
episodic buffer in terms of objects to be lower than the capacity limit for single features.
While, under a low attentional load induced by the concurrent task, WM spans for verbal and
spatial features were six and four respectively, only three cross-domain associations could be
maintained. Under medium and high attentional loads, this limit dropped to 2.9 and 2.5. In the
control experiment, applying the medium cognitive load condition, participants were not able
to maintain more than 2.3 cross-domain associations. This even lower capacity limit in the
control experiment was probably due to a higher stress on the importance of the processing
task, as participants received feedback about their performance after each trial. Overall, this
capacity limit for cross-domain associations was always lower than the capacity limit for
single features and did not exceed three objects. This result is novel as no other study had yet
given rise to clear maintenance capacity estimates for cross-domain associations.
Several previous studies have aimed at measuring the maintenance capacity limits of
associations formed within the same domain, with divergent estimates of this limit as a result.
For example, using a change-detection paradigm in which participants have to detect whether
something has changed in an array previously studied, Luck and Vogel (1997) found a
capacity limit for multi-feature visual objects of four. Surprisingly, this capacity limit proved
to be the same as for the maintenance of single features. Wheeler and Treisman (2002)
attempted to replicate some of the results of Luck and Vogel (1997) but remarkably failed in
this objective. Using the same paradigm as these authors, they found lower maintenance
33
performance for multi-feature objects than for each of their components. However, changing
the recognition paradigm introduced by Luck and Vogel (1997) to the use of a single probe
instead of the whole test array gave rise to an intermediate pattern of results. Recognition
accuracy was lower for bound visuo-spatial features than for the best-remembered feature, but
equal to the less remembered feature. Capacity limits in the recognition of objects seemed to
be constrained by the less remembered of their constituents. Allen et al. (2006) observed this
same pattern for the maintenance of separate colors and shapes as compared to bound color-
shape objects on several occasions. However, a methodological change of allowing repetition
of features within a trial altered the results to a lower recognition rate of bound objects as
compared with either one of the single features. The same phenomenon was observed by
Cowan, Blume, and Saults (2013) who studied memory for color-shape associations. The
number of items in WM for these combinations was surprisingly low, only slightly more than
one. However, when incomplete objects were taken into account (only color or shape), the
number of maintained items was as high as single-feature maintenance (i.e., about 3 items).
This same idea has been applied to the maintenance of verbal information (Cowan et al.,
2012). Chunks composed of different numbers of words (either one, two, or three) were
learned prior to testing. About three “chunks” could be maintained, while these chunks were
often incomplete. Cowan et al. (2012) described this phenomenon as chunk decomposition.
Chunks can fall apart in their different components and in that case occupy more than one slot
within WM. According to this principle, it is very reasonable to accept that less objects than
single features can be maintained.
Our results concerning memory for cross-domain associations were more in line with
Allen et al. (2006) and Cowan et al. (2012) for within-domain combinations and revealed a
lower capacity for objects than single features. Cowan et al.’s idea of chunk decomposition
could explain the results obtained in the present study and as such apply to the maintenance
34
capacity limit of the episodic buffer. However, our aim was to study the capacity of the
episodic buffer as a cross-domain system of maintenance. Because it is unclear whether
isolated features are maintained in the episodic buffer or in domain-specific peripheral
systems, we restrain our estimation of the episodic buffer capacity to the number of integrated
objects recalled. Our estimates indicate that this number is quite low and varies between 2 and
3.
The notion of chunk decomposition would nevertheless be able to explain the parallel
decline rate of single features and cross-domain associations as a function of attentional
availability. The results of the present study showed the recall of verbal features, spatial
features as well as cross-domain associations to decline to the same extent as less attentional
resources were available. This result makes sense if one assumes, following Cowan (2001),
each cross-domain association to occupy one of a fixed number of slots available in WM. The
decrease in attentional availabilty would then act in the same way on all slots resulting in a
poorer recall, independently of their content. One could however imagine that the more filled
slots would require more attention for their maintenance. Our results do not provide evidence
for this, as this would have resulted in a steeper decline for the cross-domain associations than
for the verbal or spatial feature maintenance. Whilst the often cited study of Wheeler and
Treisman (2002) claimed an additional need for attention to maintain visuo-spatial features in
an integrated format, this issue has subsequently often been questioned with most studies
agreeing on the fact that attention is not more necessary for the maintenance of bound
information than for the maintenance of single features (e.g., Allen et al., 2006; Johnson et al.
2008; Morey & Bieler, 2013). Remarkably, one of the rare studies showing the need for
attention to maintain multi-feature objects was a study on the maintenance of verbal-spatial
associations (Elsley and Parmentier, 2009). However, two comments should be made on the
Elsley and Parmentier study. First of all, their study was based on implicit binding between
35
verbal and spatial features. As already stated in the introduction, Morey (2011) has shown that
the maintenance of intentionally bound features can give rise to a higher memory
performance than the maintenance of incidentally bound features. Second, and more
importantly, their results do not contradict our observations. As Elsley and Parmentier (2009),
the present study confirms that attention is needed to maintain cross-domain associations as
its recall diminishes with reduced attentional resources. Crucially, the present results show
that the maintenance of cross-domain associations does not depend more on attention than the
maintenance of single features. In the same way as the studies on binding concluded that
creating bindings is effortless, the maintenance of these bindings over prolonged periods does
not lead to additional costs compared with the maintenance of isolated features.
While the phenomenon of chunk decomposition might thus offer an explanation for
the observed result, there are other options that are worth being considered in accounting for
the maintenance of cross-domain associations. One could for example suppose that the default
strategy to maintain objects is to separately maintain their constituents in domain-specific
buffers. Objects could then be reconstituted on the basis of serial position information, item in
position n in one buffer being associated with the item occupying the corresponding position
in the other (Cowan et al., 2006). However, the second experiment in this study does not seem
to support this alternative. In this experiment, it was shown that the maintained cross-domain
associations are not more prone to verbal or spatial interference than they are to neutral
interference of a concurrent task. It is largely agreed on that the maintenance of verbal
features decreases in the presence of verbal interference (e.g. Jarrold, Tam, Baddeley, &
Harvey, 2011; Bayliss et al., 2003), and there is also evidence that visuo-spatial interference
leads to a decrease in the maintenance of visuo-spatial features (e.g., Farmer, Berman, &
Fletcher, 1986, Logie, 1986; Logie, Zucco, & Baddeley, 1990). If the maintenance of cross-
domain associations was to depend on the maintenance of these separate features, then we
36
should have observed a lower recall performance when combining the cross-domain storage
with the verbal and spatial processing task than when combined with the neutral processing
task. This was however not the case.
The second experiment of the current study along with its control showed that the
maintenance of cross-domain associations is less prone to domain-specific interference than
the maintenance of single features. This suggests that cross-domain maintenance is not
accomplished in domain-specific buffers, but rather in a domain-general or domain-neutral
maintenance system as the episodic buffer is presumed to be. This result is supported by the
study of Morey (2009) who showed that when participants had to maintain cross-domain
objects, the verbal information was less prone to verbal interference (articulatory suppression)
than when they only had to maintain verbal features. However Morey observed also more
verbal domain-specific interference on the maintenance of spatial locations when these were
part of a cross-domain association than when maintained in isolation. Our results did not
show evidence for any kind of domain-specific interference when cross-domain items were to
be maintained. This difference might be due to the fact that we were only interested in the
maintenance of intact cross-domain associations, as compared to overall feature maintenance.
The ensemble of results led Morey to the conclusion that cross-domain objects are probably
stored in a domain-general store, while additional information might also be stored in domain-
specific storage buffers. While this additional storage was less evident for visuo-spatial
information, it most certainly was the case for verbal information.
To summarize, the present study has shown that WM is capable of holding a limited
number of cross-domain associations. This capacity limit is however lower than the capacity
limits for single verbal or spatial features. Cross-domain associations are assumed to be
maintained as integrated objects in a domain-general buffer. Indeed, this maintenance relies
on attention, as the detrimental effect of attention-demanding concurrent tasks testifies, but
37
not more than the maintenance of single features. Moreover, this maintenance proved immune
to domain-specific interference. These findings converge towards the hypothesis of some
episodic buffer as hypothesized by Baddeley (2000) and described by Repovs and Baddeley
(2006) as a capacity-limited system dependent on central executive resources and separable
from peripheral buffers.
38
Acknowledgements
This research was supported by a grant from the Swiss National Science Foundation N°
100014_132037 to Pierre Barrouillet.
39
References
Allen, R. J., Baddeley, A. D., & Hitch, G. J. (2006). Is the binding of visual features in
working memory resource-demanding? Journal of Experimental Psychology-General,
135, 298-313. doi: 10.1037/0096-3445.135.2.298
Allen, R. J., Hitch, G. J., & Baddeley, A. D. (2009). Cross-modal binding and working
memory. Visual Cognition, 17, 83-102. doi: 10.1080/13506280802281386
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its
control processes. The psychology of learning and motivation, 2, 89-195. doi:
10.1016/S0079-7421(08)60422-3
Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press.
Baddeley, A. D. (2000). The episodic buffer: a new component of working memory? Trends
in Cognitive Sciences, 4, 417-423. doi: 10.1016/S1364-6613(00)01538-2
Baddeley, A. D. (2012). Working memory: Theories, models and controversies. Annual
Review of Psychology, 63, 1-29. doi: 10.1146/annurev-psych-120710-100422
Baddeley, A. D., Allen, R. J., & Hitch, G. J. (2010). Investigating the episodic buffer.
Psychologica Belgica, 50, 223-243.
Baddeley, A. D., Allen, R. J., & Hitch, G. J. (2011). Binding in visual working memory: The
role of the episodic buffer. Neuropsychologia, 1393-1400. doi:
10.1016/j.neuropsychologia.2010.12.042
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), Recent
advances in learning and motivation (Vol. 8, pp. 647-667). New York: Academic
Press.
Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2009). Working memory and binding in
sentence recall. Journal of Memory and Language, 61, 438-456. doi:
10.1016/j.jml.2009.05.004
40
Barrouillet, P., Bernardin, S., & Camos, V. (2004). Time constraints and resource sharing in
adults' working memory spans. Journal of Experimental Psychology-General, 133,
83-100. doi: 10.1037/0096-3445.133.1.83
Barrouillet, P., & Camos, V. (2012). As Time Goes By: Temporal constraints in working
memory. Current Directions in Psychological Science, 21, 413-419. doi:
10.1177/0963721412459513
Barrouillet, P., Portrat, S., & Camos, V. (2011). On the law relating processing to storage in
working memory. Psychological Review, 118, 175-192. doi: 10.1037/A0022324
Bayliss, D. M., Jarrold, C., Gunn, D. M., & Baddeley, A. D. (2003). The complexities of
complex span: Explaining individual differences in working memory in children and
adults. Journal of Experimental Psychology-General, 132, 71-92. doi: 10.1037/0096-
3445.132.1.71
Brener, R. (1940). An experimental investigation of memory span. Journal of Experimental
Psychology, 26, 467-482. doi: 10.1037/H0061096
Camos, V., Lagner, P., & Barrouillet, P. (2009). Two maintenance mechanisms of verbal
information in working memory. Journal of Memory and Language, 61, 457-469. doi:
10.1016/j.jml.2009.06.002
Campo, P., Poch, C., Parementier, F. B. R., Moratti, S., Elsley, J., Castellanos, N. P., …
Maestu, F. (2010). Oscillatory activity in prefrontal and posterior regions during
implicit letter-location binding. Neuroimage, 49, 2807-2815. doi:
10.1016/j.neuroimage.2009.10.024
Chen, Z. J., & Cowan, N. (2009). Core verbal working-memory capacity: The limit in words
retained without covert articulation. Quarterly Journal of Experimental Psychology,
62, 1420-1429. doi: 10.1080/17470210802453977
41
Cowan, N. (1995). Attention and memory: An integrated framework. New York: Oxford
University Press.
Cowan, N. (1999). An embedded processes model of working memory. In A. Miyake & P.
Shah (Eds.), Models of Working Memory: Mechanisms of active maintenance and
executive control (pp. 62-101). Cambridge: Cambridge University Press.
Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental
storage capacity. Behavioral and brain sciences, 24, 87-114. doi:
10.1017/S0140525X01003922
Cowan, N. (2005). Working memory capacity. Hove, UK: Psychology Press.
Cowan, N., Blume, C. L., & Saults, J. S. (2013). Attention to attributes and objects in working
memory. Journal of Experimental Psychology – Learning, Memory, and Cognition,
39, 731-747. doi: 10.1037/a0029687
Cowan, N., Rouder, J. N., Blume, C. L., & Saults, J. S. (2012). Models of verbal working
memory capacity: What does it take to make them work? Psychological Review, 119,
480-499. doi: 10.1037/A0027791
Cowan, N., Saults, J. S., & Morey, C. C. (2006). Development of working memory for verbal-
spatial associations. Journal of Memory and Language, 55, 274-289. doi:
10.1016/j.jml.2006.04.002
Delvenne, J. F., Cleeremans, A., & Laloyaux, C. (2010). Feature bindings are maintained in
visual short-term memory without sustained focused attention. Experimental
Psychology, 57, 108-116. doi: 10.1027/1618-3169/a000014
Ecker, U. K., Maybery, M., & Zimmer, H. (2013). Binding of intrinsic and extrinsic features
in working memory. Journal of Experimental Psychology-General, 142, 218-234. doi:
10.1037/a0028732
42
Elsley, J. V., & Parmentier, F. B. R. (2009). Is verbal-spatial binding in working memory
impaired by a concurrent memory load? Quarterly Journal of Experimental
Psychology, 62, 1696-1705. doi: 10.1080/17470210902811231
Farmer, E. W., Berman, J. V. F., & Fletcher, Y. L. (1986). Evidence for a visuospatial
scratch-pad in working memory. Quarterly journal of experimental psychology
Section A – Human Experimental Psychology, 38, 675-688.
Fougnie, D., & Marois, R. (2009). Attentive Tracking Disrupts Feature Binding in Visual
Working Memory. Visual cognition, 17, 48-66. doi: 10.1080/13506280802281337
Gajewski, D. A., & Brockmole, J. R. (2006). Feature bindings endure without attention:
evidence from an explicit recall task. Psychonomic Bulletin & Review, 13, 581-587.
doi: 10.3758/BF03193966
Guérard, K., Morey, C. C., Lagace, S., & Tremblay, S. (2013). Asymmetric binding in serial
memory for verbal and spatial information. Memory & Cognition, 41, 378-391. doi:
10.3758/s13421-012-0275-4
Guérard, K., Tremblay, S., & Saint-Aubin, J. (2009). Similarity and binding in memory:
Bound to be detrimental. Quarterly Journal of Experimental Psychology, 62, 26-32.
doi: 10.1080/17470210802215277
Hyman, R. (1953). Stimulus information as a determinant of reaction time. Journal of
Experimental Psychology, 45, 188-196. doi: 10.1037/h0056940
Imbo, I., Vandierendonck, A., & Vergauwe, E. (2007). The role of working memory in
carrying and borrowing. Psychological research, 71, 467 – 483. doi: 10.1007/s00426-
006-0044-8
Jarrold, C., Tam, H., Baddeley, A. D., & Harvey, C.E. (2011). How does processing affect
storage in working memory tasks? Evidence for both domain-general and domain-
43
specific effects. Journal of Experimental Psychology – Learning, Memory and
Cognition, 37, 688-705. doi: 10.1037/a0023699
Johnson, J. S., Hollingworth, A., & Luck, S. J. (2008). The role of attention in the
maintenance of feature bindings in visual short-term memory. Journal of
Experimental Psychology: Human Perception and Performance, 34, 41-55. doi:
10.1037/0096-1523.34.1.41
Kahneman, D., Treisman, A., & Gibbs, B. J. (1992). The reviewing of object files - object-
specific integration of information. Cognitive Psychology, 24, 175-219. doi:
10.1016/0010-0285(92)90007-O
Karlsen, P. J., Allen, R. J., Baddeley, A. D., & Hitch, G. J. (2010). Binding across space and
time in visual working memory. Memory & Cognition, 38, 292-303. doi:
10.3758/Mc.38.3.292
Logie, R. H. (1986). Visuo-spatial processing in working memory. Quarterly Journal of
Experimental Psychology Section a-Human Experimental Psychology, 38, 229-247.
Logie, R. H., Zucco, G. M., & Baddeley, A. D. (1990). Interference with visual short-term
memory. Acta psychologica, 75, 55-74. doi: 10.1016/0001-6918(90)90066-O
Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and
conjunctions. Nature, 390, 279-281. doi: 10.1038/36846
Masson, M. E. J. (2011). A tutorial on a practical Bayesian alternative to null-hypothesis
significance testing. Behavior Research Methods, 43, 679-690. doi: 10.3758/s13428-
010-0049-5
Merkel, J. (1885). Die zeitlichen Verhaltnisse der Willensthatigkeit [The temporal conditions
of the activities of will]. Philosophica studies., 2, 73 – 127.
44
Morey, C. C. (2009). Integrated cross-domain object storage in working memory: Evidence
from a verbal-spatial memory task. Quarterly Journal of Experimental Psychology,
62, 2235-2251. doi: 10.1080/17470210902763382
Morey, C.C. (2011). Maintaining binding in workind memory: Comparing the effects of
intentional goals and incidental affordances. Consciousness and Cognition, 20, 920-
927. doi: 10.1016/j.concog.2010.12.013.
Morey, C. C., & Bieler, M. (2013). Visual short-term memory always requires general
attention. Psychonomic Bulletin & Review, 20, 163-170. doi: 10.3758/s13423-012-
0313-z
Morey, C. C., & Cowan, N. (2004). When visual and verbal memories compete: Evidence of
cross-domain limits in working memory. Psychonomic Bulletin & Review, 11, 296-
301. doi: 10.3758/Bf03196573
Morey, C. C., & Mall, J. T. (2012). Cross-domain interference costs during concurrent verbal
and spatial serial memory tasks are asymmetric. Quarterly Journal of Experimental
Psychology, 65, 1777-1797. doi: 10.1080/17470218.2012.668555
Murdock, B. B. (1960). The distinctiveness of stimuli. Psychological Review, 67, 16-31. doi:
10.1037/h0042382
Oberauer, K., & Eichenberger, S. (2013). Visual working memory declines when more
features must be remembered for each object. Memory and Cognition.Advance online
publication. doi: 10.3758/s13421-013-0333-6
Olson, I. R., & Jiang, Y. H. (2002). Is visual short-term memory object based? Rejection of
the "strong-object" hypothesis. Perception & Psychophysics, 64, 1055-1067. doi:
10.3758/Bf03194756
45
Postma, A., & DeHaan, E. H. F. (1996). What was where? Memory for object locations.
Quarterly Journal of Experimental Psychology Section a-Human Experimental
Psychology, 49, 178-199. doi: 10.1080/027249896392856
Prabhakaran, V., Narayanan, K., Zhao, Z., & Gabrieli, J. D. E. (2000). Integration of diverse
information in working memory within the frontal lobe. Nature Neuroscience, 3, 85-
90. doi: 10.1038/71156
Raye, C. L., Johnson, M. K., Mitchell, K. J., Greene, E. J., & Johnson, M. R. (2007).
Refreshing: A minimal executive function. Cortex, 43, 135-145. doi: 10.1016/S0010-
9452(08)70451-9
Repovs, G., & Baddeley, A. (2006). The multi-component model of working memory:
Explorations in experimental cognitive psychology. Neuroscience, 139, 5-21. doi:
10.1016/j.neuroscience.2005.12.061
Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial
thinking and language processing: An individual differences approach. Journal of
Experimental Psychology-General, 125, 4-27. doi: 10.1037/0096-3445.125.1.4
Szmalec, A., Vandierendonck, A., & Kemps, E. (2005). Response selection involves
executive control: Evidence from the selective interference paradigm. Memory &
Cognition, 33, 531-541. doi: 10.3758/BF03193069
Tulving, E., & Patkau, J. E. (1962). Concurrent effects of contextual constraint and word-
frequency on immediate recall and learning of verbal material. Canadian Journal of
Psychology, 16, 83-95. doi: 10.1037/H0083231
Vergauwe, E., Barrouillet, P., & Camos, V. (2009). Visual and spatial working memory are
not that dissociated after all: A time-based resource-sharing account. Journal of
Experimental Psychology-Learning, Memory, and Cognition, 35, 1012-1028. doi:
10.1037/a0015859
46
Vergauwe, E., Barrouillet, P., & Camos, V. (2010). Do mental processes share a domain-
general resource? Psychological Science, 21, 384-390. doi:
10.1177/0956797610361340
Vergauwe, E., Camos, V., & Barrouillet, P. (2013). The impact of storage on processing:
How is information maintained in working memory? Manuscript submitted for
publication.
Vergauwe, E., Dewaele, N., Langerock, N., & Barrouillet, P. (2012). Evidence for a central
pool of general resources in working memory. Journal of Cognitive Psychology, 24,
359-366. doi: 10.1080/20445911.2011.640625
Vogel, E. K., Woodman, G. F., & Luck, S. J. (2001). Storage of features, conjunctions, and
objects in visual working memory. Journal of Experimental Psychology-Human
Perception and Performance, 27, 92-114. doi: 10.1037/0096-1523.27.1.92
Wheeler, M. E., & Treisman, M. A. (2002). Binding in short-term visual memory. Journal of
Experimantal Psychology – General, 131, 46-64. doi: 10.1037//0096-3445.131.1.48
Yeh, Y. Y., Yang, C. T., & Chiu, Y. C. (2005). Binding or prioritization: The role of selective
attention in visual short-term memory. Visual Cognition, 12, 759-799. doi:
10.1080/13506280444000490
47
Footnotes
1. Although no significant difference was observed in Experiment 2 between verbal, spatial
and neutral interference, increasing cognitive load seemed to move the results in that direction
(Figure 4). It should nonetheless be mentioned that, though the three processing tasks
involved two-choice reaction, the number of possible stimuli the choice should be made on
was 24 in both the verbal and spatial processing task, while only two in the neutral processing
task. It has been shown by Merkel (1885; cited by Hyman (1953)) that reaction times are
longer when one has to make a choice reaction time for one stimulus drawn from a pool of ten
alternatives instead of two alternatives. In Experiment 2, this has as a result that the actual
cognitive load is slightly higher when using the verbal and spatial processing task than when
the neutral processing is used. The difference in actual cognitive load between the verbal or
spatial processing task and the neutral processing task increases as our manipulation of the
cognitive load increases. While the difference in actual cognitive load between the neutral and
the verbal or spatial processing task at the low level of cognitive load might have been too
small to have an effect on recall, at higher levels of cognitive loads this can lead to small
differences in recall. However, at a statistical level, this difference is still non-significant.
48
Figure 1: The structure of the multi-component working memory model after the introduction
of the episodic buffer. From “The episodic buffer: A new component of working memory?”
by A. D. Baddeley, 2000, Trends in Cognitive Sciences, 4, p 421. Copyright 2013 by
Rightslink. Reprinted with permission.
49
Figure 2: Structure of a trial for Experiment 1
50
Figure 3: Mean recall performance as a function of maintenance domain and cognitive load in
Experiment 1. Error bars represent the standard error of the mean.
0
1
2
3
4
5
6
7
LOW MEDIUM HIGH
MEAN SPAN
COGNITIVE LOAD
VERBAL
MAINTENANCE
SPATIAL
MAINTENANCE
CROSS-DOMAIN
MAINTENANCE
51
Figure 4: Mean cross-domain recall performance as a function of processing domain and
cognitive load in Experiment 2. Error bars represent the standard error of the mean.
0
1
2
3
4
LOW MEDIUM HIGH
MEAN SPAN
COGNITIVE LOAD
VERBAL
PROCESSING
SPATIAL
PROCESSING
NEUTRAL
PROCESSING
52
Figure 5: Mean recall performance as a function of maintenance domain and processing
domain in the Control Experiment. Error bars represent the standard error of the mean.
0
1
2
3
4
5
6
VERBAL
PROCESSING SPATIAL
PROCESSING
MEAN SPAN
VERBAL
MAINTENANCE
CROSS-DOMAIN
MAINTENANCE
