ArticlePDF Available

Working memory and binding in sentence recall

Authors:

Abstract and Figures

A series of experiments explored whether chunking in short-term memory for verbal materials depends on attentionally limited executive processes. Secondary tasks were used to disrupt components of working memory and chunking was indexed by the sentence superiority effect, whereby immediate recall is better for sentences than word lists. To facilitate comparisons and maximise demands on working memory, materials were constrained by re-sampling a small set of words. Experiment 1 confirmed a reliable sentence superiority effect with constrained materials. Experiment 2 showed that secondary tasks of concurrent articulation and visual choice reaction impaired recall, but did not remove or reduce the sentence superiority effect. This was also the case with visual and verbal n-back concurrent tasks (Experiment 3), and with concurrent backward counting (Experiment 4). Backward counting did however interact with mode of presenting the memory materials, suggesting that our failure to find interactions between concurrent task and materials was not attributable to our methodology. We conclude that executive processes are not crucial for the sentence chunking advantage and we discuss implications for the episodic buffer and other theoretical accounts of working memory and chunking.
Content may be subject to copyright.
Working memory and binding in sentence recall
A.D. Baddeley
a
, G.J. Hitch
a,*
, R.J. Allen
b
a
Department of Psychology, University of York, York YO10 5DD, United Kingdom
b
Institute of Psychological Sciences, University of Leeds, Leeds, LS2 9JT, United Kingdom
article info
Article history:
Received 17 December 2008
revision received 13 May 2009
Available online 14 July 2009
Keywords:
Working memory
Binding
Episodic buffer
Sentence advantage
abstract
A series of experiments explored whether chunking in short-term memory for verbal mate-
rials depends on attentionally limited executive processes. Secondary tasks were used to
disrupt components of working memory and chunking was indexed by the sentence supe-
riority effect, whereby immediate recall is better for sentences than word lists. To facilitate
comparisons and maximise demands on working memory, materials were constrained by
re-sampling a small set of words. Experiment 1 confirmed a reliable sentence superiority
effect with constrained materials. Experiment 2 showed that secondary tasks of concurrent
articulation and visual choice reaction impaired recall, but did not remove or reduce the
sentence superiority effect. This was also the case with visual and verbal n-back concurrent
tasks (Experiment 3), and with concurrent backward counting (Experiment 4). Backward
counting did however interact with mode of presenting the memory materials, suggesting
that our failure to find interactions between concurrent task and materials was not attrib-
utable to our methodology. We conclude that executive processes are not crucial for the
sentence chunking advantage and we discuss implications for the episodic buffer and other
theoretical accounts of working memory and chunking.
Ó2009 Elsevier Inc. All rights reserved.
Introduction
The modular, multi-store view of human memory raises
important and unresolved questions about how different
subsystems interact to enable the coherent operation of
the system as a whole. A central problem is how informa-
tion held in working memory interacts with more perma-
nent long-term knowledge. One aspect concerns the role
of working memory in acquiring new knowledge. For
example, phonological short-term storage appears to be
important in learning new word forms but not new associ-
ations between familiar words (Baddeley, Gathercole, &
Papagno, 1998), suggesting some degree of specialisation
in the links from working memory to long-term memory.
A different aspect concerns effects of previously acquired
knowledge on the operation of working memory. For
example, short-term memory span is greatly expanded
when the input can be recoded into higher-order units or
chunks using previously acquired knowledge (Miller,
1956). In the present paper we focus on short-term mem-
ory for sentences, which can be two or three times better
than for random words (Brener, 1940), thus far outweigh-
ing the capacity limits normally associated with working
memory. We investigate how working memory takes
advantage of long-term knowledge of linguistic constraints
on words and their order in sentences and in particular,
whether forming chunks depends on executive processes.
The theoretical significance of Brener’s (1940) finding
only began to emerge a decade later with the development
of information theory by Shannon and Weaver (1949), and
its application to immediate memory. Miller and Selfridge
(1950) used Shannon’s guessing game technique to pro-
duce word sequences that ranged from random, through
first order approximations (in which words reflected their
frequency in natural language), through second and third
up to sixth order of approximation, each increasing the
similarity of the sequences to natural text. Miller and
0749-596X/$ - see front matter Ó2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.jml.2009.05.004
*Corresponding author. Fax: +44 (0)1904 433181.
E-mail address: gjh3@york.ac.uk (G.J. Hitch).
Journal of Memory and Language 61 (2009) 438–456
Contents lists available at ScienceDirect
Journal of Memory and Language
journal homepage: www.elsevier.com/locate/jml
Selfridge (1950) and later Marks and Jack (1952) found
that immediate recall increased systematically with order
of approximation.
In his classic paper, Miller (1956) offered an interpreta-
tion of these and related results in terms of his concept of
chunking. He suggested that short-term memory has a
capacity that is limited in terms of the number of chunks
it can hold. Redundant material such as text allowed sev-
eral words to be combined into a single chunk, hence
allowing more words to be recalled. Tulving and Patkau
(1962) provided further evidence for the chunking hypoth-
esis in a study that used free recall, defining a chunk as a
sequence of words recalled in the same order as their pre-
sentation. They found that, whereas number of words re-
called increased with order of approximation to English,
number of chunks remained constant.
The chunking hypothesis gained further support from
analysis of the serial recall of letter sequences. Miller, Bru-
ner, and Postman (1954) generated sequences of letters
that approximated more or less closely to English word
structure. When presented tachistoscopically they ob-
served that higher orders of approximation to English led
to more accurate report, a result they interpreted in per-
ceptual terms. Baddeley (1964) showed that the redun-
dancy effect was present even with lengthy exposure,
interpreting the effect in terms of memory rather than per-
ception. McNulty (1966) demonstrated that, as in the case
for words, the number of chunks recalled remained con-
stant, while the number of letters per chunk increased with
order of approximation. Furthermore, memory for se-
quences of unrelated consonants is strongly influenced
by phonological similarity between the names of the let-
ters, suggesting that sequences are remembered in terms
of individual subvocalised letters. However, when order
of approximation increases and vowels are included, the
correlation with phonological similarity becomes insignif-
icant (r= .05) while letter-based predictability becomes
an important factor (r= .66) (Baddeley, 1971). There is
therefore abundant evidence that sequential redundancy
within language has a powerful impact on immediate re-
call, and that it operates through the process of chunking.
Our present interest in chunking based on language
habits was motivated by the need to address some of the
limitations of the model of working memory proposed by
Baddeley and Hitch (1974; see also Baddeley, 1986).
According to this model, working memory consists of a
limited capacity attentional system (the central executive)
that interacts with temporary stores for different kinds of
information (the phonological loop and the visuo-spatial
sketchpad). The phonological loop is a store for speech-
coded information that decays in the order of 2 s but can
be refreshed by subvocal rehearsal. It provides a simple
and coherent account of the limit on memory span for ran-
dom sequences of verbal items and its sensitivity to factors
such as word length, phonemic similarity and the suppres-
sion of articulation (see Baddeley, 2007). However, despite
successfully explaining a range of phenomena in the
immediate recall of lists, the vastly superior immediate re-
call of sentences is clearly beyond the time-based capacity
of the phonological loop. In itself this is not surprising, in
that the original model of working memory did not
attempt to specify links with long-term memory. It never-
theless serves to illustrate the point that the model must
be elaborated if it is to describe how previously acquired
knowledge interacts with information in the phonological
loop in processes such as chunking.
A related problem for the Baddeley and Hitch (1974)
model of working memory was raised by reading span,
the measure of working memory capacity devised by
Daneman and Carpenter (1980). There is copious evidence
that this measure predicts individual differences in cogni-
tive performance across a range of activities from language
comprehension to reasoning and acquiring skills in logic
(Daneman & Merikle, 1996; Kyllonen & Christal, 1990;
Kyllonen & Stephens, 1990; Turner & Engle, 1989). Reading
span requires the subject to process a series of sentences
and subsequently recall the final word of each. Given that
the sentences grossly exceed the capacity of the phonolog-
ical loop, the question again arises as to how the material is
maintained within working memory. The need to provide
an explanation for these examples involving natural
language materials, taken together with a range of related
problems, prompted Baddeley (2000) to propose a fourth
component of working memory, the episodic buffer (see
Fig. 1).
The episodic buffer is assumed to be a limited capacity
store in which information from the short-term stores and
long-term memory can be integrated into episodic chunks.
It differs from long-term episodic memory in being tempo-
rary and being dependent upon attentional control by the
central executive. The buffer is assumed to be accessible
through conscious awareness, and to be multi-dimen-
sional, reflecting its integrative function of drawing to-
gether related information from the senses, from working
memory and from long-term memory. The capacity of
the buffer is assumed to be limited in terms of the number
of multi-dimensional chunks it can hold at any one time. It
accounts for the sentence superiority effect in immediate
recall in terms of the greater size of chunks for sentences
as compared with word lists.
The episodic buffer thus emphasises the need for work-
ing memory to integrate information, in contrast to the fo-
cus of our previous research which was principally
concerned with isolating and studying the subcomponents
Fig. 1. Model of working memory (Baddeley, 2000).
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 439
of the system. The proposed new component thus made
the working memory model more compatible with other
approaches such as those of Cowan (2005) and Engle
(2002) who have tended to emphasise the executive and
integrative aspects of working memory, rather than its
subsystems. However, if the episodic buffer concept is to
prove more than a convenient label for an area of igno-
rance, it is necessary to demonstrate its capacity to gener-
ate empirical data that will enrich and extend our
understanding of working memory. The experiments that
follow are part of a program that attempts to do this.
A key assumption of the revised model was that the
central executive mediates access of information from
the phonological loop and the visuo-spatial sketchpad to
the episodic buffer (see arrows marked ain Fig. 1). The
model also included a direct link between the phonological
loop and long-term linguistic knowledge, justified by the
data on learning novel word forms (Baddeley et al., 1998)
and by reciprocal effects such as the superior immediate
serial recall of words compared with non-words (Hulme,
Maughan, & Brown, 1991) (see arrow bin Fig. 1). Thus,
according to the model, memory span for sentences is
enhanced by direct interactions between language
knowledge and the phonological loop and by attention-
demanding binding processes. The combination of these
processes is assumed to result in the formation of chunks
in the episodic buffer.
Both Cowan’s ‘‘embedded processes” model (Cowan,
1995, 1999) and the episodic buffer model (Baddeley,
2000) assign an important role to the interaction between
attention and long-term memory. Cowan assumes an
attentional focus comprising the currently active represen-
tations in long-term memory, surrounded by regions that
have recently been activated but are not in the current
focus of attention. Baddeley’s (2000) episodic buffer inter-
pretation differs in assuming that the buffer comprises a
system in which representations are both stored and
manipulated; the buffer is assumed to have links to repre-
sentations in long-term memory, but also to be an active
system that allows the creative manipulation and restruc-
turing of long-term representations by interacting with the
central executive. At an empirical level however, it is
currently hard to distinguish between this and Cowan’s
embedded processes hypothesis. In both cases, the precise
question of how attention and memory interact offers an
important problem. The experiments that follow address
this issue, with the aim of identifying constraints that
may help the further development of these models, neither
of which is currently specified in sufficient detail as to
allow unequivocal predictions to be made.
To summarise, our concern is to investigate the pro-
cesses underpinning the superior immediate serial recall
of sentences as compared to random word lists. We have
noted that sentence recall exceeds the capacity of the
phonological loop and benefits from chunking based on
linguistic knowledge. According to the revised model of
working memory (Baddeley, 2000), chunks are stored in
a limited capacity episodic buffer that requires attention
for access. Similarly, Cowan (2005) assumes that attention
is required to form new links in working memory in the
process of chunking. Thus both accounts suggest that a
secondary task requiring general attention should disrupt
the chunking process and hence remove or at the very least
reduce the sentence superiority effect. Support for the
alternative hypothesis that chunking does not require
attention would imply that these theoretical accounts re-
quire substantial revision.
We should perhaps make it clear that at the present
stage we are not attempting to analyse the psycholinguis-
tic processes such as syntax and semantics that underpin
sentence comprehension. Interesting and important
though these processes are, we assume that the role of
working memory in chunking can be tackled as a general
question without becoming embroiled in the microanalysis
of sentence processing. We begin by discussing the prob-
lem of how to minimise the role of long-term semantic
and episodic memory in immediate verbatim recall of sen-
tences, so as to focus on the role of sequential redundancy
in working memory. We propose as a tool, a new measure
we term constrained sentence span. This involves a compar-
ison between the retention of constrained sentences and
the same words presented in random order. Dual task
methods are then used to assess the extent to which the
binding advantage conferred by the sentence form is
dependent upon the various components of working
memory.
Constrained sentence span
The fact that span for natural sentences is much greater
than for arbitrary word lists gives rise to problems of scal-
ing when comparing the effects of experimental manipula-
tions on these two types of materials. Moreover, memory
for natural sentences is likely to vary greatly depending
on sentence form and content and the knowledge base of
the individual. We therefore decided to try to create a par-
adigm in which language-based sequential redundancy
played an important role, but where sentence length and
the variable contribution from semantic memory would
be better controlled. We achieved this by making our sen-
tence task much more like the standard memory span pro-
cedure in which a small number of items are repeatedly
used in different orders. We assume that, by making the
set of items readily available, we would emphasise the need
to rely on order information. Furthermore, by selecting a set
of items that were constantly re-used, we assumed that
pro-active interference would minimise the contribution
of long-term episodic and semantic memory, forcing partic-
ipants to rely upon temporary binding in working memory
to underpin their recall. Experiment 1 tested the effective-
ness of this material in reducing the difference between
memory for sentences and non-sentential word sequences
while maintaining a clear sentences advantage.
Experiment 1
Method
Participants
Twelve undergraduate and postgraduate students from
the University of York took part, receiving course credit or
440 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
payment. There were five males and seven females, with
an age range of 21–36 years.
Materials
A single set of words was used to construct both sen-
tences and lists. This consisted of 15 words, of which there
were four adjectives (tall, sad, old, fat), four nouns (teacher,
soldier, waiter, bishop), four verbs (meets, insults, follows,
helps), and three function words (not, and, or). Function
words were present in both constrained sentences and
lists, and were classed as test words both in terms of se-
quence length and recall scoring. Each word was digitally
recorded by a male English speaker for auditory
presentation. Words in both conditions were presented
individually, thus removing cues from intonation or
co-articulation present in continuous speech.
A group of 16 sequences was created for each of the
constrained sentence and list sets. Each sequence consisted
of eight words (three nouns, three adjectives, one verb, one
function word). The sentences were modelled on a tele-
graphic structure so that, while not strictly grammatically
correct, they did have clear meaning and internal redun-
dancy (e.g. tall soldier follows waiter and old sad teacher).
Lists were constructed by adjusting the order of the words
in each sentence by at least two levels of approximation
(e.g. soldier tall waiter teacher follows and old sad), to ensure
minimal meaning and redundancy.
Design and procedure
Each participant was tested in a single session lasting
approximately 15 min. Presentation of stimuli was
performed on a Macintosh computer, using a SuperCard
program. Sentences and lists were presented in two
counterbalanced blocks of two practice trials and 14 test
trials.
Participants pressed the space bar to begin presentation
of each sequence, following a 2 s delay. Auditory presenta-
tion occurred through headphones at a rate of one word
per second, using the same digital sound files for sentences
and lists. As soon as each sequence ended, participants
attempted to recall as many words as they could in serial
order. Responses were recorded on a Sony Mini-disc.
Results
Recall of each word was scored as correct if it was pro-
duced in the appropriate position relative to an adjacent
recalled word. Absolute serial position was only taken into
account for first and last words in a sequence; these were
scored as correct if produced in those positions in the recall
sequence. For example, if the words ABCDE were recalled
as CBADE, only words D and E would be scored as correct
(as, in the presented sequence, the first word was not C,
B did not directly follow C, and A did not follow B). If the
sequence BCADE was instead recalled, all words except A
would be scored as correct (as in this case, B and C were re-
called in their original relative order).
The mean proportion of words correctly recalled as a
function of serial position is plotted in Fig. 2.A28 ANOVA
revealed significant effects of item set, F(1, 11) = 76.97,
MSE = .04, p< .001,
g
2
p
¼:88, serial position, F(7, 77) =
26.99, MSE = .02, p< .001,
g
2
p
¼:71, and the item set by se-
rial position interaction, F(7, 77) = 12.9, MSE = .01,
p< .001,
g
2
p
¼:54. Thus, there was a highly significant
advantage for sentences over lists, particularly at serial
positions 3–7. Further analyses (not reported here for brev-
ity) showed that the difference between conditions was lar-
gely due to order errors.
Discussion
The results show a large sentence superiority effect de-
spite the use of constrained materials and the absence of
prosodic cues and co-articulation. However, from a meth-
odological perspective, using the same length of sequence
for the two conditions clearly risks floor and/or ceiling ef-
fects when combined with a demanding secondary task.
For that reason, we opted in subsequent experiments to at-
tempt to match overall probability of recall by reducing the
sequence length for unrelated lists.
We began by using a memory span procedure to gain a
rough estimate as to likely performance on three types of
material. One comprised our constrained sentences, a sec-
ond involved the same words presented in random order,
while a third involved sentences of increasing length se-
lected from newspapers. As expected, span for these natu-
ral or open sentences proved to be highly variable,
presumably reflecting the wide range of difference in con-
tent and syntactic structure. Overall, however, our pilot
study suggested that we might obtain a roughly equal
probability of correct sequence recall if we chose random
lists of six words, constrained sentences of eight words,
and natural sentences of 16 words.
Experiment 2 combined immediate retention of each of
these three item sets with concurrent tasks designed to
disrupt different components of working memory. While
accepting that problems might arise in comparing 16-item
open sentences with six-item word lists, the experiment
did allow us to test our assumption that our constrained
sentences were qualitatively similar to more naturalistic
material. In order to enhance the ecological validity of
our constrained sentences, we moved away from the
somewhat artificial telegraphic structure of those used in
Experiment 1 by including more function words. We also
made our word lists more list-like by not allowing them
to contain function words (excluding function words from
the scoring of sentence recall in order to maintain the
matching of items across sets).
Finally, in order to gain some information on the prob-
able contribution of gist-based episodic long-term mem-
ory, we included a delayed recognition test in which
participants attempted to identify sequences they had pre-
viously encountered. We predicted that delayed recogni-
tion would be substantially greater for open sentences
than for lists, and hoped that our constrained sentences
would resemble lists in this regard, rather than open sen-
tences, suggesting a much-reduced role of semantic gist.
Our study thus focused on the immediate recall of open
or natural sentences, constrained sentences, and lists com-
prising an arbitrary re-ordering of the content words used
in constrained sentences. We varied sequence length
across materials in an attempt to equate probability of
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 441
correct recall and studied the impact on recall and error
patterns of disrupting working memory using a secondary
task procedure. The secondary tasks were (i) articulatory
suppression, assumed to disrupt the phonological loop
while making minimal demands on attention, (ii) a
demanding four-choice continuous reaction time task
based on that used successfully by Craik, Govoni, Naveh
Benjamin, and Anderson (1996), assumed to disrupt the
central executive and the visuo-spatial sketchpad, and
(iii) the combination of suppression and choice reaction
time.
We predicted that all three concurrent task conditions
would interfere with memory for each type of material,
consistent with involvement of the phonological loop and
the executive in the immediate recall task. We assumed
that participants would form chunks for lists as well as
sentences, the principal difference being that chunks
would be larger, multi-item units for sentences. At issue
was whether concurrent tasks loading the executive would
reduce or even remove the sentence advantage in recall.
The hypothesis that chunking processes depend on con-
trolled attention predicts that an executive load will cause
more items to be lost in sentence than list recall because of
the difference in presumed chunk size for the two types of
materials.
Experiment 2
Method
Participants
Twenty-four undergraduate and postgraduate students
from the University of York took part, receiving course
credit or payment. There were seven males and 17 females,
with an age range of 18–34 years.
Materials
Four groups of eight sequences were created for each of
the open sentence, constrained sentence, and constrained
list item sets, one for each of four concurrent task
conditions. Open sentences were drawn from the
www.guardian.co.uk and www.independent.co.uk online
news resources. They consisted of 16 test words, plus a
selection of definite and indefinite articles ‘the’, ‘a’, ‘an’,
prepositions ‘of’, ‘at’, ‘in’, ‘for’, ‘on’, ‘by’, ‘to’, and the conjunc-
tion ‘and’. Open sentences were selected to sample a broad
range of commonly known subject matter. Constrained
sentences and constrained lists were constructed from a
limited pool of twelve nouns, twelve adjectives, four verbs,
and four adverbs. All words were of medium–high Kucera
and Francis (1967) frequency (mean 199.55). Constrained
sentences contained eight test words comprising three
nouns, three adjectives, one verb and one adverb. As for
open sentences, they also contained function words that
were not included in the word count. Constrained lists
were six words long, and had a similar distribution of
nouns, adjectives, verbs, and adverbs to the constrained
sentences but were non-sensical, lacked function words,
and had no syntactic or semantic structure. Examples of
each of the three item sets are provided in Table 1 with de-
tails of the sentence generation procedure given in the
Appendix. All sequences were recorded digitally in a male
English speaking voice using natural intonation.
For the delayed recognition test, two sequences were
selected from each group of eight. One of the two was used
as a ‘true’ item, while the other was altered and served as a
‘false’ item. Open sentences were altered by exchanging
between one and three content words for new words, in
order to change the underlying meaning of the sentence.
Constrained sentences and lists were altered by replacing
two content words with other experimental words from
the same class (i.e. noun for noun, verb for verb, etc.). This
procedure provided four true and four false items for each
of the three item sets, distributed evenly across the four
concurrent task conditions.
Design and procedure
Each participant was tested in a single session lasting
approximately 50 min. The main part of the experiment
examined prose and word list recall under different
Fig. 2. Correct recall as a function of serial position for sentences and lists in Experiment 1.
442 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
concurrent task conditions. A 4 3 repeated measures de-
sign was used, manipulating concurrent task condition
(undivided attention; articulatory suppression; visuo-spa-
tial continuous reaction time task; suppression + visual
task) and item set (open sentences; constrained sentences;
constrained lists). Auditory presentation of memory stim-
uli and presentation and response recording in the visuo-
spatial task was performed on a Macintosh computer,
using a program written in HyperCard.
Pre-test of visuo-spatial task. A measure of baseline per-
formance in the visuo-spatial task was obtained by giving
25 practice trials followed by 200 test trials. In each trial,
four squares each measuring 6 6 cm were presented hor-
izontally across a white background, with a black circle
(diameter 1 cm) in the centre of one of these squares. Par-
ticipants responded by pressing one of four keys on the
keyboard (‘z’, ‘x’, ‘.’, ‘/’), corresponding to the square in
which the target circle was positioned. Immediately upon
the participant’s response, the visual display was reset
and the circle appeared in one of the three remaining
squares, selected at random. The computer recorded reac-
tion times and error rates, and participants were instructed
to be both as quick and as accurate as possible. No feed-
back was provided at any point.
Memory for verbal sequences. Completion of the visuo-
spatial pre-test was followed by the central part of the
experiment. Conditions were blocked so that eight se-
quences from each of the three item sets were successively
presented under each concurrent task. The same blocking
was repeated for each concurrent task condition in turn.
The orders of both concurrent task condition and item
set were counterbalanced using Latin square designs. The
four groups of sequences for each type were used in the
same order for all participants, ensuring that each group
featured in each concurrent task condition an equal num-
ber of times.
Undivided attention (No task). Each trial commenced
with participants pressing the space bar. After a 1 s delay,
the memory stimulus was presented through speakers
positioned on either side of the computer. Participants
were instructed to listen to the words and upon comple-
tion repeat them back in the order in which they were pre-
sented. Responses were recorded using a Sony minidisk
recorder.
Articulatory suppression (AS). The same procedure was
followed with the addition that participants repeatedly
articulated ‘1-2-3-4’ from the point of pressing the space
bar to the end of stimulus presentation. They were in-
structed to speak at a constant rate demonstrated by the
experimenter (approximately two utterances per second),
and at a suitably modest volume to ensure that perception
of the auditory input was not disrupted.
Visuo-spatial task. The visuo-spatial continuous reaction
time task was implemented as described earlier. Partici-
pants performed four trials and the fourth response trig-
gered presentation of the memory sequence. They were
instructed to continue responding in the visual task as
quickly and as accurately as possible throughout presenta-
tion and recall of the memory sequence and to halt
immediately upon completion of recall. As in the other
conditions, participants began each trial at their own pace
by pressing the ‘’ key (to reset the experimental program).
Reaction times and error rates from the visual task were
recorded, and coded by the computer as having occurred
during the introductory trials, memory stimulus presenta-
tion, or recall.
Articulatory suppression + visuo-spatial task (AS + V). The
final condition was a combination of those previously
described. Participants listened to and repeated back a
sentence or list while simultaneously performing the
visuo-spatial task. In addition, they were also required to
repeatedly articulate the sequence ‘1-2-3-4’ from the first
visual trial through to completion of sentence/list presen-
tation. Thus, participants performed both suppression
and the visual task during presentation of the memory
stimuli, with only the latter also continuing through recall.
Post-test of visuo-spatial task. A second baseline measure
of performance on the visual task was obtained following
completion of all the verbal recall conditions. This was a
repeat of the pre-test with the 25 practice trials removed,
thus leaving 200 test trials. Pre- and post-test data were
then combined to provide an overall estimate of baseline
performance.
Delayed recognition test. In the final segment of the
experiment, participants were given a two-page A4 booklet
containing 24 sentences or word lists. Twelve of these
were used in the experiment, with the remaining twelve
being other experimental items with a few constituent
words altered (see Materials). The same 24 items were pre-
sented to all participants, in the same randomised order.
Participants were instructed to read each item and record
a yes/no decision as to whether it had been presented in
the experiment in the exact form displayed together with
a confidence rating of 1–5.
Results and discussion
Sentence and word list recall
Recall was scored using the procedure described in
Experiment 1, with the modification that function words
were ignored. For example, for the sentence LUCY the
OLD PILOT RAPIDLY BORROWED the SMALL RED BOOK, a
response of ‘‘Lucy the wealthy pilot rapidly stole the red
old book” would score four for the italicised items. The
mean number of words correctly recalled, collapsed across
concurrent tasks, was 9.88 (SD = 1.77) for open sentences,
6.74 (SD = .57) for constrained sentences, and 4.82
(SD = .62) for lists. Thus, many more words were correctly
recalled for open sentences than for constrained sentences,
t(23) = 10.52, p< .001, effect size (Cohen’s d) = 1.79, or for
lists, t(23) = 16.65, p< .001, (d= 3.07), with absolute recall
Table 1
Experiment 2: examples of sequences used in each item set.
Example of items
Open sentence CAR HEADLIGHTS THAT CAN HELP MOTORISTS SEE
ROUND CORNERS WILL FINALLY BE INTRODUCED
SOMETIME in the NEXT YEAR
Constrained
sentence
LUCY the OLD PILOT RAPIDLY BORROWED the
SMALL RED BOOK
Constrained List JOHN CAR INSTANTLY LARGE BORROWED WHITE
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 443
also higher for constrained sentences than for lists,
t(23) = 31.98, p< .001, (d= 3.20). Having confirmed our
assumption that natural sentences would be easier to re-
call than constrained sentences, which would in turn be
easier than lists, analysis of the effects of the concurrent
tasks focused on proportion correct, taking advantage of
our attempt to match for overall item difficulty by varying
sequence length.
The mean proportions of words correctly recalled for
each item set and concurrent task condition are displayed
in Fig. 3.A43 repeated measures ANOVA on probability
of correct recall showed a significant effect of item set,
F(2, 46) = 108.88, MSE = 01, p< .001,
g
2
p
¼:83, indicating
that we had underestimated the difficulty of the open sen-
tences. The concurrent task effect was also significant,
F(3, 69) = 41.26, MSE = .01, p< .001,
g
2
p
¼:64, but the inter-
action between item set and concurrent task was not,
F(5, 115) = 1.61, MSE = .003, p= .149,
g
2
p
¼:07. In separate
analyses of each of the concurrent tasks, the effect of vi-
suo-spatial task, F(1, 23) = 11.97, MSE = 1.92
03
,p< .01,
g
2
p
¼:34, and articulatory suppression, F(1, 23) = 77.16,
MSE = 3.58
03
,p< .001,
g
2
p
¼:77, were both significant.
Furthermore, the interaction between articulatory sup-
pression and item set was significant, F(2, 46) = 3.73,
MSE = 1.87
03
,p< .05,
g
2
p
¼:14, reflecting a slightly larger
effect of suppression on the two constrained item sets, rel-
ative to open sentences. However, a significant interaction
was not observed between item set and visuo-spatial task,
F(2, 46) = .54, MSE = 1.35
03
,
g
2
p
¼:02. Thus there were
clear interference effects in the various dual-task condi-
tions but, importantly, none of them had a selective effect
on memory for sentences as compared with memory for
lists.
Error analysis
If, as we suggest, our constrained sentences gain their
advantage from the same processes as benefit natural sen-
tence recall, then we might expect this to be reflected in
the pattern of errors. Despite being closer in span size to
lists, constrained sentences should show closer similarities
to the longer natural sentences.
Errors were categorised as order errors in which items
from the sequence were reported in the incorrect position,
omissions, intra-experiment intrusions, and semantic er-
rors. The distribution of these various error forms is shown
in Table 2, with order errors also displayed in Fig. 4. Deri-
vational, phonological and extra-experimental intrusions
were noted, but were considered too infrequent for analy-
sis, with rates of <.03 in all conditions.
Order errors. To avoid confounding differences in order
memory with differences in levels of overall item memory,
the rate of order errors is reported as a proportion of the
total number of words recalled, regardless of order (Mur-
dock, 1976). A 4 3 repeated measure ANOVA revealed
significant effects of item set, F(2, 46) = 27.29, MSE =
3.76
03
,p< .001,
g
2
p
¼:54, concurrent task condition,
F(3, 69) = 5.45, MSE = 9.66
04
,p< .01,
g
2
p
¼:19, and a sig-
nificant set by task interaction, F(6, 138) = 2.97, MSE =
8.90
04
,p< .01,
g
2
p
¼:11.
As Fig. 4 suggests, both open and constrained sentences
showed a lower proportion of order errors than lists, for
open sentences, t(23) = 4.78, p< .001, (d= 1.10), and for
constrained sentences, t(23) = 5.99, p< .001, (d= 1.17),
with order errors being slightly more frequent for open
than constrained sentences, t(23) = 2.44, p< .05, (d= .45).
This pattern suggests that retention of order is facilitated
by language-based redundancy, with this effect being at
least as great for the constrained as for the open sentences,
hence supporting the view that the constrained and open
sentences reflect a similar serial order mechanism. The
task by item set interaction reflects a clear effect of the
two conditions involving articulatory suppression that is
most marked for the lists, suggesting that the capacity
for subvocal rehearsal is particularly important in the ab-
sence of language-based sequential redundancy.
Omissions. These comprise instances when a presented
item is absent from recall, without an alternative response
being made. As shown in Table 2, omissions are
Fig. 3. Mean proportion of words correctly recalled in each item set and concurrent task condition in Experiment 2 (AS = Articulatory Suppression).
444 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
particularly prominent for open sentences. However, as
Fig. 3 shows, proportional recall of sentences was relatively
low, suggesting that this observation should be interpreted
with caution. Analysis indicated an overall effect of item
set, F(2, 46) = 137.12, MSE = .01, p< .001,
g
2
p
¼:86, and con-
current task, F(3,69) = 11.67, MSE = 3.50
03
,p< .001,
g
2
p
¼:34, but no significant interaction, F(6, 138) = 1.27,
MSE = 2.06
03
,p= .278,
g
2
p
¼:05. The main effect of concur-
rent task again appears to indicate principally a reduction
of performance under articulatory suppression, although
in this case the effect is broadly similar across conditions.
Semantic errors. These comprise intrusions with a simi-
lar meaning but distinct phonology to the target (e.g. police
for detectives, or removed for withdrawn). As Table 2
shows, semantic errors are far more common in the case
of open sentences. This could reflect a greater reliance on
semantic gist with open sentences, but might also occur
because the limited item set in the constrained conditions
allow subjects to edit out such errors from their responses.
ANOVA indicated a significant effect of item set,
F(2, 46) = 113.91, MSE = 1.20
03
p< .001,
g
2
p
¼:83, and a
small effect of concurrent task, F(3, 69) = 3.37, MSE =
2.83
04
p< .05,
g
2
p
¼:13, which again seems to reflect a
slightly higher error rate under concurrent articulatory
suppression.
Intra-experimental intrusions. These comprise intrusions
from sequences presented previously in the experiment.
They were slightly more frequent for constrained sen-
tences than open sentences or lists, particularly for the
suppression and visual + suppression conditions. ANOVA
indicated a significant effect of item set, F(2, 46) = 23.49,
MSE = 2.50
03
p< .001,
g
2
p
¼:51, concurrent task,
F(3, 69) = 3.69, MSE = 1.09
03
p< .001,
g
2
p
¼:35, and the
interaction, F(6, 138) = 4.99, MSE = 7.63
04
p< .001,
g
2
p
¼:18. Again, this interaction is difficult to interpret
due to the generally low rate of this error type.
Finally, it is notable that while articulatory suppression
clearly impaired performance, the concurrent visual task
had little effect on either the frequency or pattern of errors,
suggesting that the effects observed were mainly attribut-
able to disruption of the phonological loop. Overall, the er-
ror analysis supports the proposal that constrained
sentences, although much shorter, have much in common
with open sentences particularly in respect to the impact
of redundancy on order errors, while the limited response
set tends to reduce the frequency of semantic errors. A
similar pattern of errors was also found in our subsequent
experiments, but in the interests of brevity these data are
not reported.
Visuo-spatial task performance
Performance on the visuo-spatial task is shown in Table
3. Outliers were removed from the reaction time data by
excluding any points over three standard deviations above
the mean in a given condition. Reaction times were col-
lapsed across pre- and post-test sessions, resulting in a
mean reaction time of 387 ms (SD = 44), and an error rate
of 5.6% (SD = 3.86). Visuo-spatial task latencies, when
Table 2
Experiment 2: mean error rates (and standard deviations) in each item set
and concurrent task condition. (IEI = Intra-experimental Intrusions;
CRT = Choice Reaction Time; AS = Articulatory Suppression).
Order Omissions IEI Semantic
Open sentences
No task .05 (.05) .21 (.08) .01 (.01) .06 (.04)
Visual CRT .05 (.03) .23 (.10) .01 (.01) .06 (.03)
AS .05 (.02) .25 (.10) .01 (.01) .07 (.04)
Visual CRT + AS .06 (.05) .26 (.11) .01 (.01) .08 (.04)
Constrained sentences
No task .04 (.02) .03 (.03) .03 (.03) .00 (.00)
Visual CRT .04 (.01) .03 (.04) .04 (.03) .00 (.01)
AS .04 (.01) .09 (.07) .06 (.04) .00 (.01)
Visual CRT + AS .05 (.03) .06 (.05) .09 (.08) .00 (.01)
Constrained lists
No task .09 (.06) .02 (.03) .01 (.02) . 00 (.00)
Visual CRT .09 (.06) .03 (.04) .02 (.03) .00 (.00)
AS .11 (.07) .07 (.07) .04 (.03) .00 (.00)
Visual CRT + AS .13(.07) .07 (.10) .04 (.06) .00 (.01)
Fig. 4. Proportional order error rates for each item set and concurrent task condition in Experiment 2 (AS = Articulatory Suppression).
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 445
performed as a concurrent task are displayed in Table 3.It
is apparent that, relative to the mean baseline latency,
reaction time was slowed when encoding the to-be-
remembered sequences, and was even more impaired dur-
ing recall. Articulatory suppression slowed reaction time
further, during both encoding and recall, with one excep-
tion, namely in the open sentence recall condition. A
322 repeated measures ANOVA confirmed these
observations, with significant effects of item set,
F(2, 46) = 16.94, MSE = 2560, p< .001,
g
2
p
¼:42, suppres-
sion, F(1, 23) = 11.21, MSE = 3029, p< .01,
g
2
p
¼:33, and
task stage, F(1, 23) = 153.03, MSE = 3854, p< .001,
g
2
p
¼:87.
This pattern of results is basically as expected based on
studies of LTM by Craik et al. (1996) and Naveh-Benjamin,
Kilb, and Fisher (2006) who found effects on a concurrent
RT task to be greater during retrieval than encoding. Fur-
ther slowing occurs when an additional task, articulatory
suppression, is required. These effects do not differ be-
tween list and constrained sentence recall, with the one
anomaly occurring for the retrieval phase with open
sentences. It is unclear whether this reflects the greater
sentence length, or possibly a different pattern when gist
becomes important.
Delayed recognition of sentences and lists
This final task was included to throw light on the epi-
sodic LTM contribution to performance across the three
sets of items. Non-parametric A’ was used as the recogni-
tion measure; mean scores were .90 (SD = .09) for open
sentences, .50 (SD = .28) for constrained sentences, and
.56 (SD = .28) for constrained lists. Recognition was signif-
icantly higher for open sentences than either constrained
sentences, t(23) = 7.30, p< .001, (d= 1.73), or lists,
t(23) = 5.62, p< .001, (d= 1.61), but did not significantly
differ between the latter two sets, t(23) = .65. Indeed, fur-
ther analysis revealed that delayed recognition was not
significantly above chance (.50) on constrained sentences
or lists, t(23) = .03, p= .976, and t(23) = 1.05, p= .306,
respectively.
While these results should be interpreted with caution,
they are consistent with our initial assumption that con-
strained sentence recall would be much less likely to reflect
a major contribution from episodic LTM than would open
sentences. This may well reflect pro-active interference
from successive sequences of highly similar material, an ef-
fect that is marked in the Peterson and Peterson (1959)
short-term forgetting paradigm (Keppel & Underwood,
1962; Wickens, 1970).
To summarise, Experiment 2 showed that performance
was greatest for the open sentences and least for the lists,
with constrained sentence recall at an intermediate level,
as would be expected given chunking based on language
habits. There was a small but significant effect of the visual
RT task and a more substantial effect of articulatory sup-
pression. However, these effects did not consistently inter-
act with material type, suggesting that the factors
responsible for the advantage attributable to sentential
redundancy were not influenced differentially by the de-
mand imposed by our simultaneous tasks.
The pattern of errors for the open sentences showed a
preponderance of omissions, together with a significant
number of semantic intrusions, presumably reflecting a
reliance on gist in this condition. Such errors were extre-
mely rare in the other conditions, possibly reflecting a
shallower level of semantic coding. The contrast in the final
recognition performance between the good recognition of
open sentences and chance performance on the con-
strained and random sets is consistent with this interpre-
tation. The use of a closed word pool for constrained
sentences and lists may also have enabled the exclusion
of extra-experimental phonological forms from recall.
Overall, our results offer further support for the use of
constrained sentences as a tool for analysing the effects
of language-based redundancy on serial recall. Constrained
sentences once again show a clear advantage over lists,
indicating a robust effect of sentential redundancy in this
case when sequence lengths were unequal and chosen to
give broadly similar proportions of correct recall. However,
given that level of performance on constrained sentences
and lists in baseline conditions was approximately 90%, it
could be argued that our results may have been
constrained by ceiling effects. In addition, our selection of
concurrent tasks did not manipulate the load on different
components of working memory very systematically.
Experiment 3 therefore used a more systematic design,
employing an adaptation of the n-back technique (Jonides
et al., 1997) to devise a balanced set of four secondary
tasks. The n-back method involves tracking a series of
stimuli at a lag specified by the parameter n, allowing us
to manipulate size of lag and type of stimulus orthogonally
in a 2 2 design. Thus, we tested at 0-back (essentially,
simple shadowing) and 2-back (shadowing at a lag of
two items), and we did this separately for verbal and vi-
suo-spatial stimulus sequences. Two-back has been shown
to be a highly demanding executive task that neuroimag-
ing studies have found to depend on both the frontal re-
gions associated with executive control, and the more
posterior cortical regions associated with storage (Owen,
McMillan, Laird, & Bullmore, 2005; Smith & Jonides,
1999). Using the n-back as a concurrent task allows a
Table 3
Experiment 2: mean visual task reaction times (and standard deviations) in visual only and visual + AS conditions, during presentation and recall of verbal
sequences (AS = Articulatory Suppression).
Visual only Visual + AS
Presentation Recall Presentation Recall
Open sentences 424.09 (48.29) 568.60 (96.12) 447.15 (62.48) 559.71 (112.28)
Constrained sentences 415.46 (53.86) 481.41 (69.40) 442.58 (66.75) 504.79 (77.65)
Constrained lists 411.79 (47.07) 486.39 (77.09) 440.29 (62.56) 523.50 (83.90)
446 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
separation of the functioning of the three components of
the original working memory model, with the difference
between baseline and 0-back conditions reflecting the im-
pact of the phonological and visuo-spatial subsystems, and
the difference between 0- and 2-back conditions indicating
the further influence of the central executive. We used the
recall variant of the n-back task, which Kane, Conway,
Miura, and Colflesh (2007) have indicated as being a more
appropriate method of tapping working memory process-
ing than recognition n-back.
In this and subsequent studies we used only con-
strained sentences and constrained word lists. This choice
avoids the large differences in levels of performance asso-
ciated with open sentences, allowing us to detect a clear
effect of redundancy based on language habits without
introducing complexity due to sequence length, varied
semantic content and an additional substantial contribu-
tion from episodic LTM.
As before we expected firstly a general pattern of sec-
ondary task interference common to sentences and lists,
reflecting the role of working memory in immediate serial
recall of verbal materials. Thus, memory for both types of
materials should depend principally on the phonological
loop and the central executive and much less so the vi-
suo-spatial sketchpad, generating predictions that verbal
n-back would more interfering than visuo-spatial n-back,
and 2-back more than 0-back. As before, we expected sen-
tences to be recalled better than lists and we were inter-
ested in whether loading the central executive would
remove or reduce this sentence superiority effect.
Experiment 3
This experiment compared immediate serial recall of
constrained sentences and random word lists under base-
line conditions and combined with 0-and 2-back versions
of our verbal and visuo-spatial tasks.
Method
Participants
Twenty-four undergraduate and postgraduate students
from the University of York took part, receiving course
credit or payment. There were 10 males and 14 females,
with an age range of 18–31 years.
Materials
Five sets of 10 sequences were created for each of the
constrained sentence and constrained list sets, using the
same word pool as Experiment 2. Sentences and lists were
constructed in the same way as before, with the exception
that they were nine and seven test words long respectively
(as opposed to eight and six), to avoid possible ceiling
effects.
Design and procedure
Each participant was tested in a single session lasting
approximately 1 h. Stimulus presentation and recording
of concurrent task responses was performed on a
Macintosh computer, using a HyperCard program. Each
session started with 60 practice trials on each of the four
n-back tasks (visuo-spatial and verbal 0- and 2-back), per-
formed in a counterbalanced order, before moving on to
the sequence recall task.
Visuo-spatial n-back.A33 grid was presented on
screen, each grid square measuring 4 4 cm. On each trial,
a 1 cm diameter black circle appeared in one of the nine
locations and remained present until a response was made
or the trial timed out (after 3 s). The next trial followed
immediately, with the circle appearing in one of the eight
remaining locations, chosen at random. In the 0-back con-
dition, participants had to respond by pressing a key on the
33 numerical keypad corresponding to the on-screen
location presently occupied. In the 2-back condition, the
task was to respond to the location occupied two trials pre-
viously. In this condition, participants were presented with
two preparatory trials before each practice and test block,
to enable a 2-back response on the first trial proper. In the
preparatory trials, the word ‘Wait’ appeared below the
33 grid. Participants were instructed to let each of these
trials time out and not respond. On the third trial, they at-
tempted to respond with the location occupied in trial one,
and so on for subsequent trials.
Verbal n-back. On each trial a random digit was pre-
sented at the centre of the screen in 30 pt. Arial bold font.
In the 0-back task, participants were instructed to read
aloud the presented number, while in the 2-back task, they
were to articulate the number presented two trials previ-
ously. Again, two extra ‘Wait’ trials were added before each
block, with participants not responding until the third trial.
As soon as a verbal response was produced, the experi-
menter pressed a key, moving the program on to the next
trial and immediately prompting presentation of the next
digit.
Sequence recall. Each of the five sets of ten sentences and
lists were used equally often in each concurrent task con-
dition. These conditions were blocked so that each concur-
rent task was performed on both sentences and lists,
before moving on to the next task. The order of item set
and concurrent task conditions was counterbalanced
across the experiment, with the exception of the baseline
(undivided attention) condition. Five sentences and five
lists were presented for recall under baseline conditions
at the start of the memory phase, with the remaining five
from each set presented following completion of the con-
current task conditions, to allow an estimation of any prac-
tice effects. Each of the four concurrent task conditions was
presented in between the two halves of the baseline condi-
tion (visuo-spatial 0-back/2-back, and verbal 0-back/2-
back), with 10 sentences and 10 lists for each.
On each trial in the baseline condition, a mouse click by
the participant triggered presentation of the verbal se-
quence through headphones, following a 2-s delay. As soon
as the sequence had finished, participants attempted to re-
call as many words as they could in serial order, before
moving on to hear the next trial. In the remaining condi-
tions, presentation of the verbal sequence was preceded
by four concurrent task trials (plus two ‘Wait’ trials for
the 2-back conditions), with response on the fourth trial
triggering presentation. Participants were instructed to
continue performing the visuo-spatial or verbal concurrent
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 447
task as quickly and accurately as possible, while listening
to the item being presented. On completion of sentence
or list presentation the screen turned blank and recall
was attempted. Thus, no additional tasks were performed
during recall of the items. Participants clicked a mouse to
start the next set of concurrent task trials. Responses in
the primary and secondary tasks were recorded on a Sony
Mini-disc recorder.
Results
Sentence and word list recall
Recall was scored using the procedure described in
Experiment 2. As expected, sentential redundancy had a
beneficial impact on the mean numbers of words recalled.
In the baseline condition these were 7.20 (SD = 1.04) for
sentences and 5.18 (1.22) for lists, a highly significant dif-
ference, t(23) = 14.25, p< .001 (d= 1.74). The mean propor-
tions of words correctly recalled for each item set and
concurrent task condition are displayed in Fig. 5.A25
repeated measures analysis of variance on proportion cor-
rect revealed significant effects of item set, F(1, 23) = 43.65,
MSE = 01, p< .001,
g
2
p
¼:66, and concurrent task,
F(4, 92) = 84.52, MSE = 01, p< .001,
g
2
p
¼:79, but impor-
tantly, no interaction, F(4, 92) = .71, MSE = .01. Further
analyses revealed that recall in the visuo-spatial 0-back
condition was significantly poorer than in the baseline
condition, F(1, 23) = 10.30, MSE = 3.08
03
,p< .01,
g
2
p
¼:31,
but significantly better than the visuo-spatial 2-back con-
dition, F(1, 23) = 8.20, MSE = .01, p< .01,
g
2
p
¼:26. Recall
in the verbal 0-back condition was also significantly poorer
than in the baseline condition, F(1, 23) = 175.52,
MSE = 3.66
03
,p< .001,
g
2
p
¼:88, and significantly better
than in the verbal 2-back condition, F(1, 23) = 45.32,
MSE = .02, p< .001,
g
2
p
¼:66.
Error analyses
The different error types were scored and analysed
using the method described in Experiment 2. Derivational,
semantic, phonological, and extra-experimental intrusions
all occurred at proportional rates of less than .01 and so
were insufficiently frequent for analysis.
ANOVA on the rate of order errors revealed significant
effects of item set, F(1, 23) = 93.58, MSE = .01, p< .001,
g
2
p
¼:81, and concurrent task, F(4, 92) = 34.92, MSE =
3.56
03
,p< .001,
g
2
p
¼:35, while crucially, the set by task
interaction was not significant, F(4, 92) = 1.17, MSE =
2.85
03
,p= .328,
g
2
p
¼:03 again suggesting that the nature
of secondary task disruption was not qualitatively different
for sentential material.
Visuo-spatial and verbal n-back performance
Concurrent n-back performance was based only on tri-
als recorded during sentence or list presentation (i.e.,
excluding the four pre-presentation trials). Error rates
and mean reaction times are displayed in Table 4.
As expected, accuracy was substantially poorer in the 2-
back versions of each concurrent task. Moreover, there
were no differences associated with materials, and no
interactions between materials and concurrent task condi-
tions. However, while reaction times were faster in the vi-
sual than verbal n-back tasks, accuracy tended to be lower.
We return to this latter observation in the discussion.
Discussion
We again obtained a clear advantage of constrained
sentences over lists, and a marked overall effect of concur-
rent task. Considering first the 0-back tasks which are as-
sumed to load principally on the visuo-spatial and
phonological storage systems with minimal executive de-
mand, we found a small effect of the visuo-spatial task
and a much larger effect of articulatory suppression. In
the 2-back tasks, where more substantial storage and exec-
utive processing is required, further impairment occurred,
particularly in the case of verbal 2-back condition. How-
ever, the absence of significant item set by concurrent task
interactions indicates that a range of manipulations that
have a marked effect on performance influence sentences
and word lists to an equivalent extent, suggesting once
Fig. 5. Mean proportion of words correctly recalled in each item set and concurrent task condition in Experiment 3 (VS = Visuo-spatial).
448 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
again that sentential redundancy effects are not dependent
on the contribution of working memory to encoding.
The overall pattern of errors mirrors that observed for
proportion correct in the case of omissions, order errors
and intrusions. As in Experiment 2, however, order errors
were more frequent in lists than sentences consistent with
the assumption that sentence-based redundancy enhances
the storage of order information.
Performance on the secondary tasks was virtually per-
fect in the 0-back conditions. The visual 2-back task
showed an error rate of approximately 20%, whereas the
error rate for the verbal equivalent was around 14% when
encoding sentences or lists (see Table 4). This pattern has
implications for interpreting the verbal memory data
where, it may be recalled, concurrent verbal activity was
more disruptive than was visuo-spatial. This could con-
ceivably reflect a differential trade-off with subjects re-
quired to combine sentence memory with the visual task
tending to bias attention in the direction of memory, while
those performing the verbal task might conceivably have
shown an opposite trade-off. However, of more immediate
significance is the lack of interaction between speed or
accuracy and type of memory material, suggesting that
whatever the nature of the impact of secondary tasks on
performance, it is broadly equivalent for sentences and
lists.
Experiment 4
Our final study was concerned to test the generality of
our previous observations by extending them to visual pre-
sentation of the memory materials. It seemed possible that
reading, being a less automatic way of processing language
than hearing, might be more likely to show differential dis-
ruption under concurrent task load. Our final study there-
fore compared retention of visually and verbally presented
sentences and lists. In order to avoid peripheral visual
interference with the visual memory task, we switched
from concurrent tasks involving ongoing stimulus presen-
tation to a more traditional pair of concurrent tasks,
namely counting backwards to disrupt both phonological
and executive processing, coupled with articulatory sup-
pression to disrupt phonological alone (Allen, Baddeley, &
Hitch, 2006; Peterson & Peterson, 1959).
Visual presentation in the absence of subvocal rehearsal
inevitably removes any presentational cues from co-artic-
ulation or prosody. We therefore reverted to the materials
and method used in Experiment 1 in which we initially re-
corded all spoken words independently and presented
them in a telegraphic form including function words in
word lists as well as sentences. Experiment 4 was therefore
an attempt to replicate our previous findings with auditory
presentation and explore their generalisation to visual
presentation, using more tightly constrained stimulus
material and modified concurrent tasks Thus, for auditory
presentation we expected counting backwards to cause
additional interference over and above articulatory sup-
pression and to do so to the same extent when recalling
sentences and lists. This outcome would indicate that the
presence of intonation cues was not responsible for partic-
ipants’ ability to chunk sentences without requiring
controlled attention in Experiments 2 and 3. Assuming that
reading is more demanding of executive processes than
listening, we expected the additional interference associ-
ated with counting backwards to be greater for visual than
we had found with auditory presentation. Moreover, if the
chunking processes underlying sentence recall are
especially demanding of executive processes in the case
of visual presentation, then counting could have a bigger
effect on memory for visual sentences than visual lists.
Method
Participants
There were 32 participants (11 male and 21 female) in
this experiment. All were undergraduate or postgraduate
students at the University of York, taking part for financial
payment or course credit.
Materials
Stimulus sets and scoring were the same as in Experi-
ment 1. Four sets of sentences and four sets of lists were
created, with 11 sequences (one practice and 10 test) in
each set. Sentences consisted of seven words (two adjec-
tives, three nouns, one verb, one function), for example, tall
soldier follows waiter not old teacher, while lists were five
words (one adjective, two nouns, one verb, one function)
in length, e.g. follows old not waiter teacher. As in Experi-
ments 2 and 3, sequence lengths were chosen so as to try
and match difficulty in terms of proportion of items cor-
rectly recalled. Lists were constructed by removing a noun
and an adjective from each sentence, and adjusting the or-
der of the words by at least two levels of approximation, to
ensure minimal meaning and redundancy.
Design and Procedure
A222 repeated measures design was imple-
mented, manipulating item set (sentences, lists), modality
Table 4
Experiment 3: mean reaction times and error rates (and standard deviations) in verbal and visuo-spatial n-back, during presentation of sentences and lists.
Verbal Visuo-spatial
0-Back 2-Back 0-Back 2-Back
Reaction time (ms)
Sentences 897.15 (92.92) 860.18 (243.95) 693.21 (113.33) 623.56 (356.06)
Lists 903.33 (132.65) 863.14 (216.64) 692.10 (107.20) 591.33 (279.50)
Error rate
Sentences .00 (00) .13 (.13) .02 (.03) .20 (.15)
Lists .00 (01) .14 (.12) .02 (.05) .21 (.16)
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 449
(auditory, visual), and concurrent task (articulatory sup-
pression, backward counting). There were therefore eight
experimental blocks in total, alternating between sentence
and list blocks and grouping by modality, and at the higher
level, by concurrent task. In other words, all suppression
(or backward counting) conditions were performed to-
gether, with modality conditions grouped within these
blocks. Counterbalancing was applied at each level of
grouping. There was one practice trial and 10 test trials
within each individual sequence block. Each condition
implemented a different set of sequences (counterbal-
anced across the experiment), with sequences drawn from
a set in a random order for each participant.
The experimental session began with a baseline mea-
sure of backward counting performance. Participants were
presented with three digits by the experimenter (e.g. ‘‘2-7-
6”), and, treating it as a three-digit number, attempted to
count backwards from there (e.g. ‘‘two-seven-five, two-se-
ven-four, two-seven-three...
1
). There were five practice
trials at the start of the experiment, each lasting 8 s, with an-
other five trials at the end of the experimental session (to
control for practice effects).
Participants pressed the space bar to begin presentation
of each sequence, following a 2 s delay. Auditory presenta-
tion occurred through headphones at a rate of one word
per second, using the same digital sound files for sentences
and lists. With visual presentation, words were presented
serially at upper centre screen in black 18-pt lower case
Arial, at a rate of one per second. Following auditory or vi-
sual sequence offset, there was a 1 s blank screen delay, a
further 1 s delay containing a visual mask (subtending a vi-
sual angle of approximately 19°by 6.5°), and then the se-
rial recall phase.
In the articulatory suppression conditions, participants
repeated the digit sequence ‘‘1-2-3” out loud during se-
quence presentation, from the point of pressing the space
bar through to presentation of the visual mask. For back-
ward counting, a three-digit number was presented for
2 s at upper centre screen in black 18-pt Arial at the start
of the trial, with counting proceeding from this point
through to mask presentation.
Results
Sentence and word list recall
Recall was scored using the procedure described earlier.
As before, structure and redundancy had a beneficial im-
pact on recall. The mean number of words recalled in the
articulatory suppression condition was, with auditory pre-
sentation, 5.0 (SD = 1.1) for sentences and 3.5 (0.8) for lists,
and with visual presentation, 4.1 (SD = 1.1) for sentences
and 3.1 (1.0) for lists. Hence the sentence superiority effect
for number of words recalled was significant, F(1, 31) =
98.13, MSE = .43, p< .001,
g
2
p
¼:76. There was a significant
effect of presentational modality, F(1, 31) = 60.41,
MSE = .78, p< .001,
g
2
p
¼:66, but no interaction between
modality and materials, F(1, 31) = 2.71, MSE = .59,
p= .110,
g
2
p
¼:08. Thus, the recall advantage for sentences
over lists remains even with serial visual presentation of
words.
As previously, our principal analysis was based on mean
proportion of words correct. Recall scores for each item set
and concurrent task condition are displayed in Fig. 6.A
222 repeated measures analysis of variance revealed
no effect of item set, F(1, 31) = 0.18, MSE = 01, reflecting
successful titration of sequence lengths in matching for
difficulty level. There were significant effects of concurrent
task, F(1, 31) = 378.35, MSE = 01, p< .001,
g
2
p
¼:92, and
modality, F(1, 31) = 58.74, MSE = 01, p< .001,
g
2
p
¼:66,
with as expected, worse performance under backward
counting and with visual presentation. In addition, there
was a significant interaction between concurrent task
and modality, F(1, 31) = 4.87, MSE = 01, p< .05,
g
2
p
¼:14,
indicating a larger impact of executive load on encoding
during visual presentation. However, the concurrent task
by item set interaction was not significant, F(1, 31) = 1.25,
MSE = 01, p= .27,
g
2
p
¼:04. Thus, backward counting had
a much greater effect on recall than did suppression, but
such disruption was no greater for sentences than for lists.
The modality by item set and three-way interactions were
not significant (F< 1).
Articulatory suppression and backward counting performance
The total number of counts achieved in each trial was
divided by the length of the trial in seconds (to allow for
the different trial lengths used for sentences and lists), to
obtain a measure of the mean number of counts per sec-
ond. The mean suppression and backward counting scores
for each concurrent task condition, along with the baseline
score in backward counting, are displayed in Table 5.
A222 repeated measures analysis of variance on
the concurrent task conditions revealed significant effects
of task type, F(1, 31) = 86.74, MSE = .01, p< .001,
g
2
p
¼:74,
and sequence type, F(1, 31) = 63.21, MSE = .01, p< .001,
g
2
p
¼:67. Backward counting was performed more slowly
than articulatory suppression, consistent with our assump-
tion that counting is a more demanding task. In addition,
both concurrent tasks were slower with sentences relative
to lists. The main effect of modality was not significant,
F(1, 31) = .46, MSE = .01, though there was a significant
interaction between modality and task type, F(1, 31) =
4.21, MSE = .01, p< .05,
g
2
p
¼:12, with backward counting
being slightly slower with visual, relative to auditory pre-
sentation of sequences. There were no other significant
interactions (F< 1). Finally, performance of backward
counting as a concurrent task was compared with the base-
line measure of counting performance. A series of planned
comparisons revealed that backward counting was signifi-
cantly slower, relative to baseline during auditory sentence
presentation, t(31) = 4.84, p< .001, visual sentence presen-
tation, t(31) = 6.04, p< .001, and visual list presentation,
t(31) = 2.78, p< .01, while the effect for auditory list pre-
sentation was not significant, t(31) = 1.57, p= .128.
Discussion
Results for the auditory presentation condition are dis-
cussed first before moving on to the comparison between
1
Pilot work with backward counting in decrements of 2 and 3 led to
floor effects in some conditions.
450 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
the auditory and visual conditions. The purpose of the
auditory condition was to see whether the effects of a
demanding concurrent task found in Experiments 2 and 3
would replicate when stimuli were presented using dis-
continuous rather than continuous speech. The outcome
was straightforward. Once again sentences were recalled
better than word lists; there was substantial dual-task
interference from an attention-demanding concurrent
task; and the amount of interference was no greater for
sentences than for lists. Thus we can be confident that this
set of observations reflects the sequential redundancy of
language and that the earlier results were not due to per-
ceptual cues associated with continuous speech. Analyses
of secondary task performance confirmed that the pattern
of dual-task decrements in recall was not due to differen-
tial trade-offs between the primary and secondary tasks.
It is interesting to note that secondary task performance
was significantly poorer in the sentence than list condi-
tions. However, the effect was the same for articulatory
suppression and backward counting, suggesting that sen-
tences placed greater demands on the phonological loop
(involved in both secondary tasks) but not the central
executive. Whether this effect was due to the difference
in structure or length of the two types of sequence is
impossible to say as the two were confounded.
The pattern of results for visual presentation was virtu-
ally identical to the auditory data, except that recall was
generally poorer and backward counting caused signifi-
cantly more interference relative to articulatory suppres-
sion. The increase in dual-task interference was reflected
in slower backward counting performance with visual as
compared with auditory presentation of sequences. We ex-
pected visual presentation to be more attention-demand-
ing than auditory presentation given that participants
had to direct their attention to the stimuli and carry out
the extra process of phonological recoding. We were more
particularly concerned to see whether the sentence superi-
ority effect for visual sequences was vulnerable to an
attention-demanding concurrent task. The results were
clear-cut in showing a sentence superiority effect of the
same magnitude as with auditory presentation that re-
mained intact despite concurrent backwards counting.
Although not shown here, our data once again indicated
that sentences were associated with a reduction in order
errors, and that serial position curves were bow-shaped
for sentences as well as lists. These observations provide
further evidence that the results of Experiments 2 and 3
were not an artefact of co-articulation or prosodic speech
cues. The results thus add to our earlier evidence that
working memory takes advantage of the structural con-
straints in sentences without making demands on execu-
tive processes. It seems, therefore, that the chunking
process triggered by sequential redundancy in language
is an automatic rather than a controlled process.
General discussion
To recapitulate, we have examined the effects of
sequential redundancy in immediate serial recall by ana-
lysing the sentence superiority effect, whereby memory
for sentences is enhanced relative to memory for word
lists. We assume that memory for sentences benefits from
a process in which short-term storage of the phonological
input interacts with knowledge of the sequential redun-
dancy of language to bind together groups of items into
larger chunks. Our main aim was to explore whether the
Fig. 6. Mean proportion of words correctly recalled in each item set and concurrent task condition in Experiment 4.
Table 5
Experiment 4: Articulatory suppression and backward counting perfor-
mance in mean no. counts per second (with standard error).
Articulatory
suppression
Backward Counting
Auditory Visual Auditory Visual No sequence
Sentence 1.18 (.04) 1.19 (.05) .83 (.03) .79 (.03)
List 1.25 (.05) 1.27 (.06) .92 (.03) .89 (.03)
Baseline .96 (.02)
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 451
integration of these different types of information involves
a multi-modal episodic buffer that operates under central
executive control (Baddeley, 2000). We approached this
question by comparing the effects of different concurrent
tasks on the immediate serial recall of sentences and word
lists. As formulated, the episodic buffer hypothesis predicts
that a concurrent task taxing the limited resources of the
executive should interact with materials, thus reducing
or even removing the sentence superiority effect.
An initial methodological problem for our approach was
that memory for natural sentences is typically very much
better than memory for lists. Marked differences in perfor-
mance levels make experimental comparisons difficult,
especially as regards the detection and interpretation of
interactions. We attempted to overcome this difficulty by
developing a constrained sentence span task that combines
the presence of linguistic redundancy with features of the
classic word span task in which items are repeatedly sam-
pled from a limited set. We found a robust and reliable
sentence superiority effect for constrained sentences in a
series of four experiments encompassing major changes
in the way information was presented (whether as contin-
uous speech, disconnected speech or visually), and minor
changes in other aspects (principally variations in se-
quence length and the role of function words).
Analyses of errors of recall shed light on the basis of the
sentence superiority effect in showing lower proportions
of order errors for both open and constrained sentences
than for word lists (see Experiment 2). The reduction in or-
der errors is consistent with our assumption that memory
for sentences benefits from the sequential redundancy
inherent in language. Open sentences tended to produce
more semantic errors than either constrained sentences
or word lists, consistent with the suggestion that they ben-
efit from a cueing effect based on memory for gist. One of
the aims of using constrained sentences was to reduce the
involvement of memory for gist and the reduction in
semantic errors suggests this aim was met. A second aim
of using constrained sentences was to increase the de-
mands on working memory by sampling from a limited
word pool and using a limited range of syntactic struc-
tures. This second aim also appears to have been met in
that a post-experimental recognition test demonstrated
substantial long-term episodic memory for open sentences
but chance performance for constrained sentences (see
Experiment 2). In addition, constrained sentences led to in-
creased pro-active interference in the form of intra-exper-
imental intrusions. We conclude therefore that constrained
sentences serve their purpose of facilitating experimental
comparisons with unstructured word lists, retaining the
crucial feature of sequential redundancy while being less
reliant on episodic LTM and semantic coding than open
sentences.
Turning to the question of whether the sentence supe-
riority effect depends on the central executive, the answer
seems clear. A remarkably consistent pattern of results was
found across a range of concurrent tasks loading on execu-
tive processes (continuous choice RT in Experiment 2, 2-
back recall in Experiment 3 and backwards counting in
Experiment 4). Furthermore, this pattern was obtained
regardless of whether the memory materials were
presented orally with intonation (as in Experiments 2
and 3), or orally without intonation or visually (as in
Experiment 4). In each case the concurrent task impaired
recall to the same extent for sentences and lists and in
no case was the sentence superiority effect removed or re-
duced by a concurrent task involving executive processes.
This pattern of effects was primarily reflected in order er-
rors, the tendency for sentences to generate fewer order
errors than lists, which we interpret as an indicator of
sequential chunking. We note also that less demanding
concurrent tasks assumed to load on the visuo-spatial
and phonological subsystems (articulatory suppression in
Experiment 2; visuo-spatial and verbal 0-back in Experi-
ment 3) gave rise to similar results. The amounts of inter-
ference were typically smaller and once again the sentence
superiority effect remained intact, implying that while
both the phonological loop and the visuo-spatial sketchpad
contribute to recall, they do this no more for sentences
than lists. Thus, our main conclusion is that the processes
involved in chunking occur more or less automatically,
with no special dependence on the central executive,
inconsistent with our initial suggestion the executive plays
a critical role in accessing the episodic buffer.
It could however be argued that the type of interaction
we were looking for was too difficult to detect because our
methods were not sufficiently sensitive. However, in the
present investigation our methodology was not only con-
sistent in indicating no differential effect of executive load
on memory for sentences and lists across a range of exper-
iments: it was also demonstrably capable of revealing this
type of interaction. Thus, in Experiment 4 we were able to
detect a significant differential effect of executive load for
visual and auditory methods of presenting the memory
materials, whilst at the same time finding no differential
effect on sentences vs. lists. Whilst this is not a conclusive
argument, it does at least indicate that our methodology
was sensitive, suggesting that any executive involvement
in chunking was at most slight.
It is important to comment on the fact that in the pres-
ent investigation we have not assessed chunking directly,
preferring instead to use the sentence superiority effect
as an index of chunking. We took this decision because
our experiments were exploring the general question of
the relationship between chunking and attention and we
sought a way of avoiding the well-known problems of
defining and measuring chunks (and indeed detailed issues
concerning language processing as discussed in the intro-
duction). However we note that if one is prepared to make
plausible assumptions about chunking it is possible to
design experiments that allow sentence and list recall to
be scored in terms of chunks as well as items (see e.g.
Gilchrist, Cowan, & Naveh-Benjamin, 2008). Adopting this
kind of approach would seem important for making further
progress in understanding where and how chunking
processes take place.
The most obvious way of accounting for our results in
terms of the concept of the episodic buffer as a temporary
store for bound information is to modify the assumption
that access to the buffer is dependent on the central
executive (Baddeley, 2000). That is, we would assume that
chunks enter the episodic buffer where they become
452 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
available to attention, but the binding processes whereby
chunks are formed and enter the buffer occur outside
attention. Thus, at any instant the episodic buffer contains
a limited number of recently formed chunks that can enter
into the focus of attention.
This view of chunking is consistent with evidence on
sentence comprehension reviewed by Caplan, Waters,
and DeDe (2008), who conclude that the ‘‘initial, auto-
matic, on-line, obligatory, unconscious, processes that as-
sign the structure and literal meaning of a sentence”
require specialised resources and do not draw on the cen-
tral executive of working memory. We would tentatively
equate the creation of chunks during sentence presenta-
tion with these ‘‘interpretive” processes, which Caplan
et al. distinguish from ‘‘post-interpretive” processes that
do require executive processes. The latter would corre-
spond in our account to the use of focused attention to ac-
tively manipulate chunks stored in the episodic buffer. Our
interpretation is also broadly consistent with the work of
Potter and colleagues who have shown that readers can
understand and remember sentences presented at rates
of up to 12 words per second whereas only very few unre-
lated words can be remembered at such fast rates. Potter
interprets these observations in terms of rapid formation
of chunks through automatic activation of knowledge in
long-term memory, and temporary storage of such chunks
in a conceptual short-term memory store (CSTM). Our no-
tion of the episodic buffer is similar to CSTM but differs in
that the buffer is assumed to be multi-modal and capable
of binding together and integrating information from other
parts of the memory system besides long-term memory,
including sensory stores, the phonological loop and the vi-
suo-spatial sketchpad.
Our interpretation is also broadly compatible with the-
oretical approaches that identify working memory with
the focus of attention (Cowan, 2005) or controlled atten-
tion (Engle, 2002) in that we assume the contents of the
episodic buffer are available to attention. However we
would argue that the contents of the buffer are not neces-
sarily in the focus of attention at any instant, an idea that
might be mapped onto Cowan’s proposal that a limited
amount of additional information lies immediately outside
the current focus of attention in a readily accessible state.
A potentially important difference between ourselves and
Cowan is our suggestion that the limit on number of
chunks in recall may be a function of the storage capacity
of the episodic buffer, rather than as Cowan suggests,
purely a limit on the capacity of focused attention. Another
distinction is that Cowan (2008) regards the phonological
loop and visuo-spatial sketchpad as part of activated
long-term memory whereas we see them as temporary
stores, separate from the episodic buffer. Seeking to clarify-
ing these key issues is an important goal for future
research.
Our present interpretation is apparently somewhat at
odds with Ericsson & Kintsch’s (1995) theory of long-term
working memory (LTWM). Ericsson & Kintsch discuss
memory for language in terms of a model in which chunks
are formed in long-term memory. They assume working
memory holds pointers to a retrieval plan that reflects
the organisation of information in long-term memory
and is used to unpack the contents of chunks during recall.
This approach stands in contrast to our assumption that
chunks are held in a temporary episodic buffer. Ericsson
and Kintsch support their view by showing that memory
for language can survive interruptions and delays under
some circumstances. They also show more generally that
this applies to memory for any type of information where
the individual has extensive knowledge, such as a chess ex-
pert’s memory for the arrangement of pieces on a chess-
board. Interestingly, our present data on post-experimen-
tal recognition memory suggest that the type of chunk
formed may be an important factor as regards its durability
in memory. Thus, long-term recognition memory was rela-
tively good for open sentences but at chance for con-
strained sentences, an observation we attributed to the
role of gist in the former but not the latter (see Experiment
2). This observation suggests a speculative way of combin-
ing the LTWM and episodic buffer approaches into a uni-
fied account using Hebb’s idea of short-term memory as
a set of transient reverberatory circuits whose persistence
results in the formation of more stable representations in
long-term memory (Hebb, 1949). We would suppose that
the episodic buffer holds chunks in the form of such cir-
cuits and that this activity is more readily overwritten
through similarity-based interference when it is primarily
syntactic (as in constrained sentences) than when it is pri-
marily semantic (as in open sentences). We would suppose
further that chunks that are not overwritten quickly have a
higher probability of being stored in long-term memory,
and that these circumstances correspond most closely to
those discussed by Ericsson and Kintsch.
Consideration of the role of long-term memory in recall
raises the further question whether participants in our
experiments might not have formed chunks at all, simply
using their knowledge that they had received a sentence
rather than a list to constrain guessing during recall. Thus,
in the middle of a recall protocol the participant would
have some idea what parts of speech could come next in
the case of a sentence, and this benefit of guessing alone
would improve memory for the order of the words.
Although such guessing would almost certainly have oc-
curred to some extent, it does not come close to explaining
the size of the sentence superiority effect we observed. For
example, in Experiment 2 participants recalled on average
about five list items and seven items from constrained sen-
tences, a difference of two items. However, participants
made on average very much less than one order error in
recalling lists. Moreover, given the size of the experimental
word pool (12 adjectives, 12 nouns, four verbs, four ad-
verbs), the average probability of converting an order error
into a correct response by constrained guessing is not high.
Thus, we conclude that the sentence superiority effect was
too big to be explained solely in terms of redintegrative
processes during recall, while accepting that such pro-
cesses no doubt made some contribution to recall
performance.
We should note at this point that we have reached a
broadly similar theoretical conclusion in the case of bind-
ing in visual working memory. The binding of features such
as shape and colour into objects is also unaffected by a
concurrent executive load (Allen et al., 2006), as is the
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 453
binding effect of symmetry in visual STM (Rossi-Arnaud,
Pieroni, & Baddeley, 2006), suggesting the presence of a di-
rect link between the visuo-spatial sketchpad and the epi-
sodic buffer. We have even found evidence that binding
shape and colour is automatic when one feature is pre-
sented visually and the author auditorily (Allen, Hitch, &
Baddeley, 2009). It could be argued that the binding of fea-
tures into objects and the detection of symmetry involve a
much more basic level of integrative processing than sen-
tential chunking based on syntactic and semantic informa-
tion. There is however evidence of automatic processing of
visual stimuli at a much deeper and more complex level
from an extensive series of studies by Potter and her asso-
ciates (e.g., Potter, 1993; Potter & Lombardi, 1990; Potter,
Straub, & O’Connor, 2004). For example, viewers can iden-
tify a complex visual target such as ‘‘two men drinking
beer” from a sequence presented as rapidly as eight items
per second (Potter, 1975). We have already noted a close
degree of correspondence between Potter’s notion of CSTM
and our view of the episodic buffer.
More generally still, it is important to note that the
present experiments only speak to chunking processes that
operate within individual sentences. Jefferies, Lambon
Ralph, and Baddeley (2004) investigated the role of differ-
ent components of working memory in chunking in a study
that required participants to recall a series of sentences or
scrambled word strings under either control conditions, or
when performing an attention-demanding continuous
four-choice reaction time task similar to that used in
Experiment 2 (Craik et al., 1996). Number of sentences or
lists presented was set at 50% above span for that type of
material, participants being given three successive learning
trials. The effect of the concurrent task was significant but
not large, and differed between the two sets of materials.
There was little initial effect on immediate recall of the
word lists, but the size of the effect grew over trials, while
for the sets of sentences, the trend was in the opposite
direction. Jefferies et al. interpreted the result as suggest-
ing that attentional resources were required for chunking
under some circumstances. In the case of the scrambled
words, this built up as subjective units developed, requir-
ing more than one trial as observed by Tulving (1966).In
the case of sentences, the principal challenge came from
linking together the unrelated sentences; once they were
chunked into a structure, the need for executive processing
became less. This interpretation was supported by a fur-
ther experiment in which a third condition was added,
comprising sentences that were thematically related,
hence likely to require minimal attention for inter-sen-
tence chunking. The concurrent task had little effect on this
condition, supporting our chunking hypothesis. Thus, it
seems that there are limits on the generality of the conclu-
sions from the present experiments, and chunking may in-
volve attention when binding in memory for super-span
verbal materials is made particularly difficult.
In conclusion, our investigation into the role of senten-
tial redundancy in binding has been guided by the multi-
component working memory framework, which operates
by framing questions rather than making precise predic-
tions. In answering such questions, we provide data that
have potential importance for a range of alternative
theoretical approaches. What then, are the implications
of our results for the concept of an episodic buffer? Firstly,
they demonstrate that despite its lack of precise predic-
tions, the theoretical framework is capable of posing
important questions concerning the role of attention in
binding in short-term memory, producing unexpected re-
sults that will need to be accounted for by any adequate
theory of binding. Secondly, they strongly suggest a modi-
fication of the initial Baddeley (2000) model, which pro-
posed that the buffer, important for binding features into
episodes, could be accessed only via the central executive.
It is now clear that such binding can occur without disrup-
tion, even when the central executive is loaded to a point
at which a substantial decrease in overall performance oc-
curs. This conclusion appears to hold not only for chunking
in verbal short-term memory but also for different kinds of
feature binding in visual short-term memory (Allen et al.,
2006, 2009; Rossi-Arnaud et al., 2006).
The earlier version of the buffer (Baddeley, 2000) could
be seen as involving a minimal change from the initial
Baddeley and Hitch (1974) concept of a central executive
that could both process and store information, replacing
the single executive with two systems that interact in a
powerful but unspecified way. Our recent results suggest
a more complete separation between executive control
and passive episodic storage. The evidence now appears
to favour direct access to the buffer through the phonolog-
ical and visuo-spatial subsystems, and possibly more di-
rectly through perception and from LTM. This would
imply a relatively passive store, with both binding and the
manipulation of bound representations depending on pro-
cesses that operate outside the buffer, the former being
automatic and the latter involving executive processes.
We plan to continue to investigate this and related hypoth-
eses in future work.
Acknowledgments
We are grateful to David Turk for help in developing the
constrained sentence test, to Elizabeth Littlewood for assis-
tance with data collection, and to the Medical Research
Council Grant G9423916 for financial support. We thank
also several anonymous reviewers and Nelson Cowan for
valuable comments on earlier versions of this paper.
Appendix A
Constrained word pool and sequence construction rules
Nouns (1) Nouns (2) Nouns (3) Verbs
Lucy pilot book gave
Peter athlete bicycle borrowed
Mary musician suit cleaned
John lawyer car stole
Adjectives (1) Adjectives (2) Adjectives (3) Adverbs
red large frightened rapidly
black small wealthy instantly
green old angry cheerfully
white new hopeful easily
454 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
Constrained sentences
a. Each sequence contained 1–2 words from each sub-
set of nouns and adjectives, one verb, and one
adverb, sampled without replacement.
b. There were also 2–4 function words from the, to, and,
from, and for, sampled with replacement. These were
not included in any data analysis.
c. A variety of grammatical templates were allowed,
within the bounds of acceptable linguistic structure.
Constrained lists
a. Each list was constructed by removing all function
words and two test words from a constrained
sentence.
b. Word order was then manipulated so that each
sequence was at least two orders of approximation
from an acceptable grammatical structure.
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.
Allen, R. J., Hitch, G. J., & Baddeley, A. D. (2009). Cross-modal binding and
working memory. Visual Cognition, 17, 83–102.
Baddeley, A. D. (1964). Immediate memory and the ‘‘Perception” of letter
sequences. Quarterly Journal of Experimental Psychology, 16, 364–367.
Baddeley, A. D. (1971). Language habits, acoustic confusability and
immediate memory for redundant letter sequences. Psychonomic
Science, 22, 120–121.
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.
Baddeley, A. D. (2007). Working memory, thought and action. Oxford:
Oxford University Press.
Baddeley, A. D., Gathercole, S. E., & Papagno, C. (1998). The phonological
loop as a language learning device. Psychological Review, 105,
158–173.
Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower
(Ed.). Recent advances in learning and motivation (Vol. 8, pp. 47–89).
New York: Academic Press.
Brener, R. (1940). An experimental investigation of memory span. Journal
of Experimental Psychology, 26, 467–483.
Caplan, D., Waters, G., & DeDe, G. (2008). Specialized verbal working
memory for language comprehension. In A. R. A. Conway, C. Jarrold,
M. J. Kane, A. Miyake, & J. Towse (Eds.), Variation in working memory
(pp. 272–302). New York, NY: Oxford University Press.
Cowan, N. (1995). Attention and memory: An integrated framework.
Oxford psychology series (Vol. 26). New York: Oxford University Press.
Cowan, N. (1999). An embedded-processes model of working memory. In
A. M. P. Shah (Ed.), Models of working memory (pp. 62–101).
Cambridge, UK: Cambridge University Press.
Cowan, N. (2005). Working memory capacity. Hove: Psychology Press.
Cowan, N. (2008). What are the differences between long-term, and
short-term, and working memory? In W. S. Sossin, J.-C. Lacaille, V. F.
Castellucci, & S. Belleville (Eds.), Progress in brain research.Essence of
memory (Vol. 169, pp. 323–338). Amsterdam: Elsevier.
Craik, F. I. M., Govoni, R., Naveh Benjamin, M., & Anderson, N. D. (1996).
The effects of divided attention on encoding and retrieval processes in
human memory. Journal of Experimental Psychology: General, 125,
159–180.
Daneman, M., & Carpenter, P. A. (1980). Individual differences in working
memory and reading. Journal of Verbal Learning and Verbal Behaviour,
19, 450–466.
Daneman, M., & Merikle, P. M. (1996). Working memory and language
comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3,
422–433.
Engle, R. W. (2002). Working memory capacity as executive attention.
Current Directions in Psychological Science, 11, 19–23.
Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory.
Psychological Review, 102, 211–245.
Gilchrist, A. L., Cowan, N., & Naveh-Benjamin, M. (2008). Working
memory capacity for spoken sentences decreases with adult ageing:
Recall of fewer, but not smaller chunks in older adults. Memory, 16,
773–787.
Hebb, D. O. (1949). The organization of behavior: A neuropsychological
theory. New York: Wiley.
Hulme, C., Maughan, S., & Brown, G. D. A. (1991). Memory for familiar and
unfamiliar words: Evidence for a long-term memory contribution to
short-term memory span. Journal of Memory and Language, 30,
685–701.
Jefferies, E., Lambon Ralph, M. A., & Baddeley, A. D. (2004). Automatic and
controlled processing in sentence recall: The role of long-term and
working memory. Journal of Memory and Language, 51, 623–
643.
Jonides, J., Schumacher, E. H., Smith, E. E., Lauber, E. J., Awh, E., Minoshima,
S., et al. (1997). Verbal working memory load affects regional brain
activation as measured by PET. Journal of Cognitive Neuroscience, 9,
462–475.
Kane, M. J., Conway, A. R. A., Miura, T. K., & Colflesh, G. J. (2007). Working
memory, attention control and the N-back task: A question of
construct validity. Journal of Experimental Psychology: Learning,
Memory and Cognition, 33, 615–622.
Keppel, G., & Underwood, B. J. (1962). Proactive inhibition in short-term
retention of single items. Journal of Verbal Learning and Verbal
Behavior, 1, 153–161.
Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day
American English. Providence, RI: Brown University Press.
Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little
more than) working memory capacity. Intelligence, 14, 389–
433.
Kyllonen, P. C., & Stephens, D. L. (1990). Cognitive abilities as
determinants of success in acquiring logic skill. Learning and
Individual Differences, 2, 129–160.
Marks, M. R., & Jack, O. (1952). Verbal context and memory span for
meaningful material. American Journal of Psychology, 65, 298–
300.
McNulty, J. A. (1966). A partial learning model of recognition memory.
Canadian Journal of Psychology, 20, 302–315.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some
limits on our capacity for processing information. Psychological
Review, 63, 81–97.
Miller, G. A., Bruner, J. S., & Postman, L. (1954). Familiarity of letter
sequences and tachistoscope identification. Journal of General
Psychology, 50, 129–139.
Miller, G. A., & Selfridge, J. A. (1950). Verbal context and the recall of
meaningful material. American Journal of Psychology, 63, 176–
185.
Murdock, B. B. (1976). Item and order information in short-term serial
memory. Journal of Experimental Psychology: General, 105, 191–
216.
Naveh-Benjamin, M., Kilb, A., & Fisher, T. (2006). Concurrent task effects
on memory encoding and retrieval: Further support for an
asymmetry. Memory & Cognition, 34, 90–101.
Owen, A. M., McMillan, K. M., Laird, A. R., & Bullmore, E. (2005). N-
Back working memory paradigm: A meta-analysis of normative
functional neuroimaging studies. Human Brain Mapping, 25,
46–59.
Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual
verbal items. Journal of Experimental Psychology, 58, 193–
198.
Potter, M. C. (1975). Meaning in visual search. Science, 187, 965–
966.
Potter, M. C. (1993). Very short-term conceptual memory. Memory &
Cognition, 21, 156–161.
Potter, M. C., & Lombardi, L. (1990). Regeneration in the short-term recall
of sentences. Journal of Memory and Language, 29, 633–654.
Potter, M. C., Straub, A., & O’Connor, D. H. (2004). Pictorial and conceptual
representation of glimpsed pictures. Journal of Experimental
Psychology: Human Perception and Performance, 30, 478–
489.
Rossi-Arnaud, C., Pieroni, L., & Baddeley, A. D. (2006). Symmetry and
binding in visuo-spatial working memory. Journal of Cognitive
Neuroscience, 139, 393–400.
Shannon, C. E., & Weaver, W. (1949). The mathematical theory of
communication. Urbana: University of Illinois Press.
Smith, E. E., & Jonides, J. (1999). Storage and executive processes in the
frontal lobes. Science, 283, 1657–1661.
A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456 455
Tulving, E. (1966). Subjective organization and effects of repetition in
multi-trial free recall learning. Journal of Verbal Learning and Verbal
Behavior, 5, 193–197.
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.
Turner, M. L., & Engle, R. W. (1989). Is working memory capacity
task-dependent? Journal of Memory and Language, 28, 127–
154.
Wickens, D. D. (1970). Encoding categories of words: An empirical
approach to meaning. Psychological Review, 77, 1–15.
456 A.D. Baddeley et al. / Journal of Memory and Language 61 (2009) 438–456
... Research on WM drawing on the integration of a large number of empirical investigations of typical and nontypical participants constituted a major source of and the catalyst for the formulation of the seminal multicomponent model by Baddeley and colleagues [10,24]. This multicomponent model of WM by Baddeley [25,26] has thus become the most widely cited framework across multiple disciplines, including language and bilingualism/SLA research [27][28][29]. ...
... In contrast to phonological WM, other WM components, such as visuospatial WM and the episodic buffer in Baddeley's model, have received much less enthusiasm among second-language researchers [22,49], nor have the linguistics-and psycholinguistics-oriented components, such as semantic WM and orthographic WM, been vigorously researched (e.g., [50]). In more recent years, though, Baddeley and colleagues [26] have explored the role of the episodic buffer in binding visual information such as objects. This emerging trend will hopefully renew momentum towards investigating the implications of visuospatial WM and the episodic buffer for sentence recall (e.g., [51,52]) and for following spoken instructions (i.e., the enactment effect) [53,54]. ...
Article
Full-text available
Although emotional or affective working memory (WM) is quite well established in general psychology, not much research has looked into its potential implications for the language sciences and bilingualism and second language acquisition (SLA) research until recently. To fill this gap, this paper aims to propose that WM has not just cognitive implications, but its affective dimension may also make complementary and unique contributions to language and bilingualism/SLA research. Towards this end, we first briefly synthesize the cognitive views of WM conceptions and assessment procedures in the current language sciences and bilingualism/SLA research. Next, we turn to discuss the theoretical models and assumptions of affective WM and explore their theoretical implications for bilingualism/SLA research based on emerging empirical evidence. Then, we propose a conceptual framework integrating cognitive and affective WM perspectives and further provide guidelines for designing affective WM span tasks that can be used in future affective WM–language research, focusing on the construction procedures of several emotion-based affective WM span tasks (e.g., the emotional reading span task, the emotional operation span task, and the emotional symmetry span task) as examples. Overall, we argue that affective feelings are also an integral part of the mental representations held in WM and future research in the language sciences and bilingualism/SLA should incorporate both cognitive and affective WM dimensions.
... In this preliminary study, we focus on sign language interpreters who, like spoken language interpreters, make use of auditory WM, but also need to rely on visuospatial WM. According to Baddeley's model of WM (Baddeley et al., 2009;Baddeley, 2012), auditory memory and visual memory are maintained through separate functional components: the phonological loop, and the visual sketchpad, respectively. This suggests that there may be different memory requirements or processes for SLI compared to spoken language interpreting, given the different modalities that must be attended to. ...
Article
Full-text available
Sign language interpreting (SLI) is a cognitively challenging task performed mostly by second language learners (i.e., not raised using a sign language as a home language). SLI students must first gain language fluency in a new visuospatial modality and then move between spoken and signed modalities as they interpret. As a result, many students plateau before reaching working fluency, and SLI training program drop-out rates are high. However, we know little about the requisite skills to become a successful interpreter: the few existing studies investigating SLI aptitude in terms of linguistic and cognitive skills lack baseline measures. Here we report a 3-year exploratory longitudinal skills assessments study with British Sign Language (BSL)-English SLI students at two universities (n = 33). Our aims were two-fold: first, to better understand the prerequisite skills that lead to successful SLI outcomes; second, to better understand how signing and interpreting skills impact other aspects of cognition. A battery of tasks was completed at four time points to assess skills, including but not limited to: multimodal and unimodal working memory, 2-dimensional and 3-dimensional mental rotation (MR), and English comprehension. Dependent measures were BSL and SLI course grades, BSL reproduction tests, and consecutive SLI tasks. Results reveal that initial BSL proficiency and 2D-MR were associated with selection for the degree program, while visuospatial working memory was linked to continuing with the program. 3D-MR improved throughout the degree, alongside some limited gains in auditory, visuospatial, and multimodal working memory tasks. Visuospatial working memory and MR were the skills closest associated with BSL and SLI outcomes, particularly those tasks involving sign language production, thus, highlighting the importance of cognition related to the visuospatial modality. These preliminary data will inform SLI training programs, from applicant selection to curriculum design.
... Information may be lost at each stage. Sensory memory is the initial stage and can contain all information, and lasts a few seconds or fractions of a second (Sperling, 1960); short-term memory refers to a memory system that stores small amounts of information over a short period of time (Baddeley et al., 2009), containing 5 to 7 items lasting for 15 to 30 seconds (Brown, 1958;Peterson, 1959); long-term memory contains a large amount of information and can last for years or even decades. Each type of memory has different properties. ...
Thesis
Full-text available
Education has been the focus of extensive debate related to design. The main reason for this controversy is information asymmetry between ‘teaching’ and ‘learning’ in design education. In other words, wording used in design education has a structure that is difficult to reproduce as images in the brains of trainees who receive them. Most design courses in design disciplines are studio-centered. In other words, a large number of participants participate in design studios and experience various shapes. The sensory meaning memory obtained from this shape experience is an implicit memory that cannot be consciously used or communicated. The purpose of this study is to identify the reasons for information asymmetry in design education and to propose ways to overcome this problem in order to have a positive impact on design education. This study focuses on the relationship between learning and memory in the context of the neural activity structure of the brain provided by neurology. First, starting from the memory structure, analyze the relationship between different memory types and learning. Afterwards, the process of memory encoding and retrieval was analyzed for its influence on learning effect. Finally, analyzing the memory styles used in existing design education, explores the impact of visual information encoded and transmitted through semantics in design on learning efficiency. Studies have shown that the information educators transmit to students is implicitly encoded in existing design education processes. However, the accuracy and communicability of content encoded through implicit memory are inevitably reduced relative to content encoded through explicit memory. Therefore, a meaningful asymmetry exists between them. Since the method of encoding visual information in language affects memory retrieval, the design knowledge generated in this way also affects its utilization. Efficiency of design education and learning can be improved by matching the encoding contexts for transmission, storage and extraction.
... In the past 20 years, working memory (WM) research has become a hot topic and frontier in cognitive neuroscience. As an essential component of executive function, working memory is essential for the function of the brain and refers to a memory system with limited capacity for temporarily storing and managing information during the performance of cognitive tasks [2,3]. Among many working memory models, Baddeley's multicomponent model is relatively mature and well-established and has attracted widespread attention from researchers [4]. ...
Article
Full-text available
(1) Background: The combined n-back + Go/NoGo paradigm was used to investigate whether audiovisual interactions interfere with or facilitate WM. (2) Methods: College students were randomly assigned to perform the working memory task based on either a single (visual or auditory) or dual (audiovisual) stimulus. Reaction times, accuracy, and WM performance were compared across the two groups to investigate effects of audiovisual interactions. (3) Results: With low cognitive load (2-back), auditory stimuli had no effect on visual working memory, whereas visual stimuli had a small effect on auditory working memory. With high cognitive load (3-back), auditory stimuli interfered (large effect size) with visual WM, and visual stimuli interfered (medium effect size) with auditory WM. (4) Conclusions: Audiovisual effects on WM follow the resource competition theory, and the cognitive load of a visual stimulus is dominated by competition; vision always interferes with audition, and audition conditionally interferes with vision. With increased visual cognitive load, competitive effects of audiovisual interactions were more obvious than those with auditory stimuli. Compared with visual stimuli, audiovisual stimuli showed significant interference only when visual cognitive load was high. With low visual cognitive load, the two stimulus components neither facilitated nor interfered with the other in accordance with a speed–accuracy trade-off.
Article
The pronounceability of visually-presented words has been suggested to influence the retention of words. However, it remains unclear as to whether pronounceability affects the encoding process of visually-presented words or whether it only influences subsequent process of memory retention. We investigated the recognition thresholds of tachistoscopically presented words. In Experiment 1, we classified words based on a prior study of word pronounceability and compare the recognition thresholds among easy-to-pronounce words, difficult-to-pronounce words, and non-words. In Experiment 2, we conducted a word naming task and classified words as easy-to-pronounce and difficult-to-pronounce based on speech duration of the words. In both experiments, no significant differences were observed in the recognition thresholds for easy-to-pronounce and difficult-to-pronounce words. These findings suggest that the recognition processing of visually-presented words is identical regardless of word pronounceability and that pronounceability may influence the subsequent process of memory retention. Moreover, we also investigate the possible application of a presentation-duration design for dynamic information.
Chapter
In this chapter main concepts from cognitive psychology are introduced in relation to driving, to provide a theoretical foundation for automotive interaction design. Insights concerning topics such as human information processing, human perception, attention, memory, learning and decision making are summarised. Going beyond cognitive psychology, we also summarise insights concerning emotions in relation to driving.
Article
Objectives: This research aimed at examining the effectiveness of the working memory (WM) intervention using melodies by comparing two groups of elderly subjects with Dementia of Alzheimer’s Type (DAT).Methods: A total of 10 patients with DAT (Global Deterioration Scale degree 6) were divided into a WM+MIT (melodic intonation therapy) group (n= 5) and a WM ONLY group (n= 5). The WM intervention consisted of 12 sessions, twice a week, for 20 minutes per day. The protocol consisted of 10 steps. At each step, the WM+MIT group was treated using melodies. The melody was developed as an MIT that takes into account the characteristics of Korean language based on shared syntactic integration resource hypothesis (SSIRH). However, the WM ONLY group was treated without applying the melodies.Results: The results revealed first that there were no significant differences in the pre- and post- performance levels between the two groups for the tasks; both for those with interventions and those for which interventions were not used. Secondly, the differences between pre- and post-performance assessments of the tasks that have interventions were found to be statistically significant within the WM+MIT and the WM ONLY group. The results of the pre- and post- assessments of the tasks that did not have interventions were not statistically significant.Conclusion: Although the results of the pre- and post- assessments of tasks that did not have interventions were not statistically significant, the performances of both groups improved in the post-assessments in general, proving the efficacy of the intervention.
Article
Objective: Short-term memory (STM) binding tests assess the ability to temporarily hold conjunctions between surface features, such as objects and their colors (i.e., feature binding condition), relative to the ability to hold the individual features (i.e., single feature condition). Impairments in performance of these tests have been considered cognitive markers of Alzheimer's disease (AD). The objective of the present study was to conduct a meta-analysis of results from STM binding tests used in the assessment of samples mapped along the AD clinical continuum. Method: We searched PubMed, Scopus, and Web of Science for articles that assessed patients with AD (from preclinical to dementia) using the STM binding tests and compared their results with those of controls. From each relevant article, we extracted the number of participants, the mean and standard deviations from single feature and of feature binding conditions. Results across studies were combined using standardized mean differences (effect sizes) to produce overall estimates of effect. Results: The feature binding condition of the STM binding showed large effects in all stages of AD. However, small sample sizes across studies, the presence of moderate to high heterogeneity and cross-sectional, case-controls designs decreased our confidence in the current evidence. Conclusions: To be considered as a cognitive marker for AD, properly powered longitudinal designs and studies that clearly relate conjunctive memory tests with biomarkers (amyloid and tau) are still needed. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
Article
Thematic and taxonomic semantic organizations in working memory (WM) performance seem to change during development but is not clear how and when. A sample of 698 participants divided in 5 groups was tested on a new semantic WM task devised to: i) manipulate semantic associations, and ii) test their effect on word recall and intrusion errors. Results indicated that WM performance is modulated by semantic LTM with clear developmental changes. Indeed, we showed that from infancy to 6;11 years, taxonomic and thematic associations exerted similar effects on verbal recall, whereas starting from 7;0 years of age onward, recall was more facilitated by taxonomic associations (than thematic associations). The results provide a more detailed picture of the interactions between semantic organization and WM performance in a developmental context.
Article
Full-text available
Research from a working memory perspective on the encoding and temporary maintenance of sequential instructions has established a consistent advantage for enacted over verbal recall. This is thought to reflect action planning for anticipated movements at the response phase. We describe five experiments investigating this, comparing verbal and enacted recall of a series of action–object pairings under different potentially disruptive concurrent task conditions, all requiring repetitive movements. A general advantage for enacted recall was observed across experiments, together with a tendency for concurrent action to impair sequence memory performance. The enacted recall advantage was reduced by concurrent action for both fine and gross concurrent movement with the degree of disruption influenced by both the complexity and the familiarity of the movement. The results are discussed in terms of an output buffer store of limited capacity capable of holding motoric plans for anticipated action.
Article
This paper is concerned with the relationship between two central constructs—reasoning ability and working-memory capacity—which arise from two distinct bodies of literature on individual differences in cognition, the psychometric and the information-processing, respectively. In four separate studies (N = 723, 412, 415, and 594), we assessed reasoning ability using various tests from the psychometric literature, and working-memory capacity using tests constructed according to Baddeley's (1986) definition of working memory. Confirmatory factor analysis yielded consistently high estimates of the correlation between working-memory capacity and reasoning ability factors (r = .80 to .90). We also found differentiation between the two factors: Reasoning correlated comparatively highly with general knowledge; working-memory capacity correlated comparatively highly with processing speed. Inspection of residuals from model fitting suggested the existence of a verbal versus quantitative content factor. We discuss the implications of our results for what they tell us about the nature of reasoning, and the nature of working memory.
Chapter
This chapter is divided into two parts. The first describes the effect of Pat Rabbitt's influence in encouraging the first author to use the increasingly sophisticated methods of ageing research to answer questions about the fundamental characteristics of working memory, together with reflections on why so little of this work reached publication. The second part presents a brief review of the literature on working memory and ageing, followed by an account of more recent work attempting to apply the traditional method of experimental dissociation to research on normal ageing and Alzheimer's disease. The discussion suggests that even such simple methods can throw light on both the processes of ageing and the understanding of working memory.
Article
The idea of one's memory "filling up" is a humorous misconception of how memory in general is thought to work; it is actually has no capacity limit. However, the idea of a "full brain" makes more sense with reference to working memory, which is the limited amount of information a person can hold temporarily in an especially accessible form for use in the completion of almost any challenging cognitive task. This groundbreaking book explains the evidence supporting Cowan's theoretical proposal about working memory capacity, and compares it to competing perspectives. Cognitive psychologists profoundly disagree on how working memory is limited: whether by the number of units that can be retained (and, if so, what kind of units and how many?), the types of interfering material, the time that has elapsed, some combination of these mechanisms, or none of them. The book assesses these hypotheses and examines explanations of why capacity limits occur, including vivid biological, cognitive, and evolutionary accounts. The book concludes with a discussion of the practical importance of capacity limits in daily life. Incorporating the latest from the recent surge in research into working memory capacity limits and the remarkable new insights provided by neuroimaging techniques, this book serves as an invaluable resource for all memory researchers and is accessible to a wide range of readers.
Article
This book is the magnum opus of one of the most influential cognitive psychologists of the past 50 years. This new volume on the model he created (with Graham Hitch) discusses the developments that have occurred in the past 20 years, and places it within a broader context. Working memory is a temporary storage system that underpins onex' capacity for coherent thought. Some 30 years ago, Baddeley and Hitch proposed a way of thinking about working memory that has proved to be both valuable and influential in its application to practical problems. This book updates the theory, discussing both the evidence in its favour, and alternative approaches. In addition, it discusses the implications of the model for understanding social and emotional behaviour, concluding with an attempt to place working memory in a broader biological and philosophical context. Inside are chapters on the phonological loop, the visuo-spatial sketchpad, the central executive and the episodic buffer. There are also chapters on the relevance to working memory of studies of the recency effect, of work based on individual differences, and of neuroimaging research. The broader implications of the concept of working memory are discussed in the chapters on social psychology, anxiety, depression, consciousness, and on the control of action. Finally, the author discusses the relevance of a concept of working memory to the classic problems of consciousness and free will.
Article
For decades, the fundamental processes underlying memory and attention have been understood within an "information processing" framework in which information passes from one processing stage to another, leading eventually to a response. More recently, however, the attempt to build a general theoretical framework for information processing has been largely supplanted in favor of two more recent approaches: mathematical models of processing and direct investigations of brain function. This book reconciles theoretical conflicts in the literature to present an important, analytical update of the traditional information-processing approach by modifying it to incorporate the last few decades of research on memory, attention, and brain functioning. Throughout, the book cogently considers and ultimately refutes recent challenges to the fundamental assumption of the existence of special short-term memory and selective attention faculties. It also draws a key distinction between memory processes operating inside and outside of the focus of attention. The book hopes to foster an understanding of how memory and attention operate together, and how both functions are produced by brain processes.