Rev

AR Further The Mind and Brain of
Click here for quick links to
Annual Reviews content online,
including:
Short-Term Memory
• Other articles in this volume
Annu. Rev. Psychol. 2008.59:193-224. Downloaded from arjournals.annualreviews.org

• Top cited articles
• Top downloaded articles John Jonides, Richard L. Lewis, Derek Evan Nee,
• AR’s comprehensive search
Cindy A. Lustig, Marc G. Berman,
and Katherine Sledge Moore
by National Taiwan University on 10/02/09. For personal use only.

Department of Psychology, University of Michigan, Ann Arbor, Michigan 48109;
email: jjonides@umich.edu

Annu. Rev. Psychol. 2008. 59:193–224 Key Words
First published online as a Review in Advance on working memory, attention, encoding, storage, retrieval
September 12, 2007

The Annual Review of Psychology is online at Abstract
http://psych.annualreviews.org
The past 10 years have brought near-revolutionary changes in psy-
This article’s doi: chological theories about short-term memory, with similarly great
10.1146/annurev.psych.59.103006.093615
advances in the neurosciences. Here, we critically examine the major
Copyright c 2008 by Annual Reviews. psychological theories (the “mind”) of short-term memory and how
All rights reserved
they relate to evidence about underlying brain mechanisms. We focus
0066-4308/08/0203-0193$20.00 on three features that must be addressed by any satisfactory theory
of short-term memory. First, we examine the evidence for the archi-
tecture of short-term memory, with special attention to questions of
capacity and how—or whether—short-term memory can be sepa-
rated from long-term memory. Second, we ask how the components
of that architecture enact processes of encoding, maintenance, and
retrieval. Third, we describe the debate over the reason about forget-
ting from short-term memory, whether interference or decay is the
cause. We close with a conceptual model tracing the representation
of a single item through a short-term memory task, describing the
biological mechanisms that might support psychological processes
on a moment-by-moment basis as an item is encoded, maintained
over a delay with some forgetting, and ultimately retrieved.

193

Contents
INTRODUCTION . . . . . . . . . . . . . . . . . 194 The Relationship of Short-Term
WHAT IS THE STRUCTURE Memory Processes
OF SHORT-TERM MEMORY? . 195 to Rehearsal . . . . . . . . . . . . . . . . . . . 206
Multistore Models that WHY DO WE FORGET? . . . . . . . . . . 207
Differentiate Short- and Decay Theories: Intuitive
Long-Term Memory . . . . . . . . . . 195 but Problematic . . . . . . . . . . . . . . . 207
Unitary-Store Models that Interference Theories:
Combine Short-Term and Comprehensive but Complex . . 210
Long-Term Memory . . . . . . . . . . 197 Interference Effects

Controversies over Capacity . . . . . . 199 in Short-Term Memory . . . . . . . . 210
Summary . . . . . . . . . . . . . . . . . . . . . . . . 201 A SUMMARY OF PRINCIPLES
WHAT PROCESSES OPERATE AND AN ILLUSTRATION OF

ON THE STORED SHORT-TERM MEMORY
INFORMATION? . . . . . . . . . . . . . . . 202 AT WORK . . . . . . . . . . . . . . . . . . . . . . 212
The Three Core Processes of Principles of Short-Term
Short-Term Memory: Memory . . . . . . . . . . . . . . . . . . . . . . 212
Encoding, Maintenance, A Sketch of Short-Term Memory
and Retrieval . . . . . . . . . . . . . . . . . . 202 at Work . . . . . . . . . . . . . . . . . . . . . . . 213
Neural Mechanisms of Short- and Postscript: Revisiting Complex
Long-Term Memory Retrieval . 205 Cognition . . . . . . . . . . . . . . . . . . . . . 216

INTRODUCTION adding “4” and “8”). While attending to local
portions of the problem, you must also keep
Mentally add 324 and 468.
accessible the parts of the problem that are
Follow the instructions to complete any
not in the current focus of attention (e.g., that
form for your federal income taxes.
you now have the units digit “2” as a portion
Read and comprehend this sentence.
of the ﬁnal answer). These tasks implicate a
short-term memory (STM). In fact, there is
What are the features of the memory system hardly a task that can be completed without
that allows us to complete these and other the involvement of STM, making it a critical
complex tasks? Consider the opening exam- component of cognition.
ple. First, you must create a temporary rep- Our review relates the psychological phe-
resentation in memory for the two numbers. nomena of STM to their underlying neural
This representation needs to survive for sev- mechanisms. The review is motivated by three
eral seconds to complete the task. You must questions that any adequate account of STM
then allocate your attention to different por- must address:
tions of the representation so that you can ap- 1. What is its structure?
ply the rules of arithmetic required by the task. A proper theory must describe an architec-
By one strategy, you need to focus attention ture for short-term storage. Candidate com-
on the “tens” digits (“2” and “6”) and miti- ponents of this architecture include storage
gate interference from the other digits (e.g., buffers, a moving and varying focus of atten-
“3” and “4”) and from partial results of previ- tion, or traces with differing levels of acti-
ous operations (e.g., the “12” that results from vation. In all cases, it is essential to provide

194 Jonides et al.

a mechanism that allows a representation to WHAT IS THE STRUCTURE
exist beyond the sensory stimulation that OF SHORT-TERM MEMORY?
caused it or the process that retrieved the rep-
resentation from long-term memory (LTM). Multistore Models that Differentiate
This architecture should be clear about its Short- and Long-Term Memory
psychological constructs. Furthermore, being In his Principles of Psychology, William James
clear about the neural mechanisms that imple- (1890) articulated the view that short-term
ment those constructs will aid in development (“primary”) memory is qualitatively different
of psychological theory, as we illustrate below. from long-term (“secondary”) memory (see
2. What processes operate on the stored also Hebb 1949). The most influential suc-
information? cessor to this view is the model of STM
developed by Baddeley and colleagues (e.g.,

A proper theory must articulate the processes Baddeley 1986, 1992; Baddeley & Hitch 1974;
that create and operate on representations. Repov & Baddeley 2006). For the years 1980
Candidate processes include encoding and to 2006, of the 16,154 papers that cited “work-

maintenance operations, rehearsal, shifts of ing memory” in their titles or abstracts, fully
attention from one part of the representation 7339 included citations to Alan Baddeley.
to another, and retrieval mechanisms. Some According to Baddeley’s model, there are
of these processes are often classified as exec- separate buffers for different forms of infor-
utive functions. mation. These buffers, in turn, are separate
3. What causes forgetting? from LTM. A verbal buffer, the phonological
loop, is assumed to hold information that can
A complete theory of STM must account be rehearsed verbally (e.g., letters, digits). A
for the facts of forgetting. Traditionally, the visuospatial sketchpad is assumed to maintain
two leading contending accounts of forget- visual information and can be further frac-
ting have relied on the concepts of decay and tionated into visual/object and spatial stores
interference. We review the behavioral and (Repov & Baddeley 2006, Smith et al. 1995).
neurophysiological evidence that has tradi- An episodic buffer that draws on the other
tionally been brought to the table to distin- buffers and LTM has been added to account
guish decay and interference accounts, and we for the retention of multimodal information
suggest a possible mechanism for short-term (Baddeley 2000). In addition to the storage
forgetting. buffers described above, a central executive
Most models of STM fall between two is proposed to organize the interplay between
extremes: Multistore models view STM and the various buffers and LTM and is implicated
LTM as architecturally separate systems that in controlled processing.
rely on distinct representations. By contrast, In short, the multistore model includes
according to unitary-store models, STM and several distinctions: (a) STM is distinct from
LTM rely largely on the same representations, LTM, (b) STM can be stratified into different
but differ by (a) the level of activation of these informational buffers based on information
representations and (b) some of the processes type, and (c) storage and executive processes
that normally act upon them. We focus on are distinguishable. Evidence in support of
the distinctions drawn by these theories as we these claims has relied on behavioral interfer-
examine the evidence concerning the three ence studies, neuropsychological studies, and
questions that motivate our review. In this dis- neuroimaging data.
cussion, we assume that a representation in
memory consists of a bundle of features that Evidence for the distinction between
define a memorandum, including the context short- and long-term memory. Studies of
in which that memorandum was encountered. brain-injured patients who show a deficit

www.annualreviews.org • The Mind and Brain of Short-Term Memory 195

in STM but not LTM or vice versa lead intervals, providing support for the earlier
to the implication that STM and LTM neuropsychological work dissociating long-
are separate systems.1 Patients with parietal and short-term memory. As we elaborate be-
and temporal lobe damage show impaired low, however, there are other possible in-
short-term phonological capabilities but in- terpretations, especially with regard to the
tact LTM (Shallice & Warrington 1970, Vallar MTL’s role in memory.
& Papagno 2002). Conversely, it is often
claimed that patients with medial temporal Evidence for separate buffers in short-
lobe (MTL) damage demonstrate impaired term memory. The idea that STM can be
LTM but preserved STM (e.g., Baddeley & parceled into information-specific buffers first
Warrington 1970, Scoville & Milner 1957; we received support from a series of studies of
reinterpret these effects below). selective interference (e.g., Brooks 1968, den

Neuroimaging data from healthy subjects Heyer & Barrett 1971). These studies relied
have yielded mixed results, however. A meta- on the logic that if two tasks use the same
analysis comparing regions activated during processing mechanisms, they should show in-

verbal LTM and STM tasks indicated a great terfering effects on one another if performed
deal of overlap in neural activation for the concurrently. This work showed a double dis-
tasks in the frontal and parietal lobes (Cabeza sociation: Verbal tasks interfered with verbal
et al. 2002, Cabeza & Nyberg 2000). Three STM but not visual STM, and visual tasks in-
studies that directly compared LTM and STM terfered with visual STM but not verbal STM,
in the same subjects did reveal some regions lending support to the idea of separable mem-
selective for each memory system (Braver ory systems (for reviews, see Baddeley 1986
et al. 2001, Cabeza et al. 2002, Talmi et al. and Baddeley & Hitch 1974).
2005). Yet, of these studies, only one found The advent of neuroimaging has allowed
that the MTL was uniquely activated for LTM researchers to investigate the neural correlates
(Talmi et al. 2005). What might account for of the reputed separability of STM buffers.
the discrepancy between the neuropsycholog- Verbal STM has been shown to rely primarily
ical and neuroimaging data? on left inferior frontal and left parietal cor-
One possibility is that neuroimaging tasks tices, spatial STM on right posterior dorsal
of STM often use longer retention intervals frontal and right parietal cortices, and ob-
than those employed for neuropsychological ject/visual STM on left inferior frontal, left
tasks, making the STM tasks more similar to parietal, and left inferior temporal cortices
LTM tasks. In fact, several studies have shown (e.g., Awh et al. 1996, Jonides et al. 1993,
that the MTL is important when retention in- Smith & Jonides 1997; see review by Wager &
tervals are longer than a few seconds (Buffalo Smith 2003). Verbal STM shows a marked left
et al. 1998, Cabeza et al. 2002, Holdstock et al. hemisphere preference, whereas spatial and
1995, Owen et al. 1995). Of the studies that object STM can be distinguished mainly by
compared STM and LTM in the same sub- a dorsal versus ventral separation in poste-
jects, only Talmi et al. (2005) used an STM rior cortices (consistent with Ungerleider &
retention interval shorter than five seconds. Haxby 1994; see Baddeley 2003 for an account
This study did find, in fact, that the MTL of the function of these regions in the service
was uniquely recruited at longer retention of STM).
The more recently postulated episodic
buffer arose from the need to account for in-
1
Another line of neural evidence about the separability of teractions between STM buffers and LTM.
short- and long-term memory comes from electrophysio- For example, the number of words recalled in
logical studies of animals engaged in short-term memory
tasks. We review this evidence and its interpretation in The an STM experiment can be greatly increased
Architecture of Unitary-Store Models section. if the words form a sentence (Baddeley et al.

196 Jonides et al.

1987). This “chunking” together of words to cortex and posterior parietal cortex (Wager &
increase short-term capacity relies on addi- Smith 2003). By contrast, storage processes
tional information from LTM that can be recruit predominately posterior areas in pri-
used to integrate the words (Baddeley 2000). mary and secondary association cortex. These
Thus, there must be some representational results corroborate the evidence from lesion
space that allows for the integration of infor- studies and support the distinction between
mation stored in the phonological loop and storage and executive processing.
LTM. This ability to integrate information
from STM and LTM is relatively preserved
even when one of these memory systems is Unitary-Store Models that Combine
damaged (Baddeley & Wilson 2002, Baddeley Short-Term and Long-Term Memory
et al. 1987). These data provide support for an The multistore models reviewed above com-

episodic buffer that is separable from other bine assumptions about the distinction be-
short-term buffers and from LTM (Baddeley tween short-term and long-term systems, the
2000, Baddeley & Wilson 2002). Although decomposition of short-term memory into

neural evidence about the possible localiza- information-specific buffers, and the separa-
tion of this buffer is thin, there is some sugges- tion of systems of storage from executive func-
tion that dorsolateral prefrontal cortex plays tions. We now consider unitary models that
a role (Prabhakaran et al. 2000, Zhang et al. reject the first assumption concerning distinct
2004). systems.

Evidence for separate storage and Contesting the idea of separate long-term
executive processes. Baddeley’s multistore and short-term systems. The key data sup-
model assumes that a collection of processes porting separable short-term and long-term
act upon the information stored in the systems come from neuropsychology. To re-
various buffers. Jointly termed the “central view, the critical contrast is between patients
executive,” these processes are assumed to be who show severely impaired LTM with ap-
separate from the storage buffers and have parently normal STM (e.g., Cave & Squire
been associated with the frontal lobes. 1992, Scoville & Milner 1957) and those who
Both lesion and neuroimaging data sup- show impaired STM with apparently nor-
port the distinction between storage and ex- mal LTM (e.g., Shallice & Warrington 1970).
ecutive processes. For example, patients with However, questions have been raised about
frontal damage have intact STM under condi- whether these neuropsychological studies do,
tions of low distraction (D’Esposito & Postle in fact, support the claim that STM and LTM
1999, 2000; Malmo 1942). However, when are separable. A central question is the role
distraction is inserted during a delay inter- of the medial temporal lobe. It is well estab-
val, thereby requiring the need for executive lished that the MTL is critical for long-term
processes to overcome interference, patients declarative memory formation and retrieval
with frontal damage show significant memory (Gabrieli et al. 1997, Squire 1992). However,
deficits (D’Esposito & Postle 1999, 2000). By is the MTL also engaged by STM tasks?
contrast, patients with left temporo-parietal Much research with amnesic patients show-
damage show deficits in phonological storage, ing preserved STM would suggest not, but
regardless of the effects of interference (Vallar Ranganath & Blumenfeld (2005) have sum-
& Baddeley 1984, Vallar & Papagno 2002). marized evidence showing that MTL is en-
Consistent with these patterns, a meta- gaged in short-term tasks (see also Ranganath
analysis of 60 functional neuroimaging stud- & D’Esposito 2005 and Nichols et al. 2006).
ies indicated that increased demand for exec- In particular, there is growing evidence
utive processing recruits dorsolateral frontal that a critical function of the MTL is to


establish representations that involve novel sense of distinct underlying neural systems.
relations. These relations may be among fea- Instead, the evidence points to a model in
tures or items, or between items and their which short-term memories consist of tempo-
context. By this view, episodic memory is a rary activations of long-term representations.
special case of such relations (e.g., relating a Such unitary models of memory have a long
list of words to the experimental context in history in cognitive psychology, with early
which the list was recently presented), and the theoretical unification achieved via interfer-
special role of the MTL concerns its binding ence theory (Postman 1961, Underwood &
capabilities, not the timescale on which it op- Schultz 1960). Empirical support came from
erates. STM that is apparently preserved in demonstrations that memories in both the
amnesic patients may thus reflect a preserved short and long term suffered from proac-
ability to maintain and retrieve information tive interference (e.g., Keppel & Underwood

that does not require novel relations or bind- 1962).
ing, in keeping with their preserved retrieval Perhaps the first formal proposal that
of remote memories consolidated before the short-term memory consists of activated long-

amnesia-inducing lesion. term representations was by Atkinson &
If this view is correct, then amnesic pa- Shiffrin (1971, but also see Hebb 1949). The
tients should show deficits in situations that idea fell somewhat out of favor during the
require STM for novel relations, which they hegemony of the Baddeley multistore model,
do (Hannula et al. 2005, Olson et al. 2006b). although it was given its first detailed com-
They also show STM deficits for novel mate- putational treatment by Anderson (1983). It
rials (e.g., Buffalo et al. 1998, Holdstock et al. has recently been revived and greatly devel-
1995, Olson et al. 1995, 2006a). As mentioned oped by Cowan (1988, 1995, 2000), McElree
above, electrophysiological and neuroimag- (2001), Oberauer (2002), Verhaeghen et al.
ing studies support the claim that the MTL (2004), Anderson et al. (2004), and others.
is active in support of short-term memories The key assumption is the construct of a very
(e.g., Miyashita & Chang 1968, Ranganath & limited focus of attention, although as we
D’Esposito 2001). Taken together, the MTL elaborate below, there are disagreements re-
appears to operate in both STM and LTM to garding the scope of the focus.
create novel representations, including novel One shared assumption of these models
bindings of items to context. is that STM consists of temporary activa-
Additional evidence for the STM-LTM tions of LTM representations or of represen-
distinction comes from patients with perisyl- tations of items that were recently perceived.
vian cortical lesions who are often claimed to The models differ from one to another re-
have selective deficits in STM (e.g., Hanley garding specifics, but Cowan’s model (e.g.,
et al. 1991, Warrington & Shallice 1969). Cowan 2000) is representative. According to
However, these deficits may be substan- this model, there is only one set of represen-
tially perceptual. For example, patients with tations of familiar material—the representa-
left perisylvian damage that results in STM tions in LTM. These representations can vary
deficits also have deficits in phonological pro- in strength of activation, where that strength
cessing in general, which suggests a deficit varies as a function of such variables as recency
that extends beyond STM per se (e.g., Martin and frequency of occurrence. Representations
1993). that have increased strength of activation are
more available for retrieval in STM experi-
The architecture of unitary-store models. ments, but they must be retrieved nonetheless
Our review leads to the conclusion that short- to participate in cognitive action. In addition,
and long-term memory are not architecturally these representations are subject to forgetting
separable systems—at least not in the strong over time. A special but limited set of these

198 Jonides et al.

representations, however, can be within the regions are clearly differentiated by infor-
focus of attention, where being within the fo- mation type (e.g., auditory, visual, spatial),
cus makes these representations immediately which could support the information-specific
available for cognitive processing. According buffers postulated by multistore models.
to this and similar models, then, STM is func- Unitary-store models focus on central capac-
tionally seen as consisting of LTM represen- ity limits, irrespective of modality, but they
tations that are either in the focus of attention do allow for separate resources (Cowan 2000)
or at a heightened level of activation. or feature components (Lange & Oberauer
These unitary-store models suggest a dif- 2005, Oberauer & Kliegl 2006) that occur at
ferent interpretation of frontal cortical in- lower levels of perception and representation.
volvement in STM from multistore models. Multi- and unitary-store models thus both
Early work showing the importance of frontal converge on the idea of modality-specific

cortex for STM, particularly that of Fuster representations (or components of those rep-
and Goldman-Rakic and colleagues, was first resentations) supported by distinct posterior
seen as support for multistore models (e.g., neural systems.

Funahashi et al. 1989, Fuster 1973, Jacobsen
1936, Wilson et al. 1993). For example, single-
unit activity in dorsolateral prefrontal cortex Controversies over Capacity
regions (principal sulcus, inferior convexity) Regardless of whether one subscribes to
that was selectively responsive to memoranda multi- or unitary-store models, the issue of
during the delay interval was interpreted as ev- how much information is stored in STM has
idence that these regions were the storage sites long been a prominent one (Miller 1956).
for STM. However, the sustained activation Multistore models explain capacity estimates
of frontal cortex during the delay period does largely as interplay between the speed with
not necessarily mean that this region is a site which information can be rehearsed and the
of STM storage. Many other regions of neo- speed with which information is forgotten
cortex also show activation that outlasts the (Baddeley 1986, 1992; Repov & Baddeley
physical presence of a stimulus and provides 2006). Several studies have measured this limit
a possible neural basis for STM representa- by demonstrating that approximately two sec-
tions (see Postle 2006). Furthermore, increas- onds worth of verbal information can be re-
ing evidence suggests that frontal activations circulated successfully (e.g., Baddeley et al.
reflect the operation of executive processes 1975).
[including those needed to keep the represen- Unitary-store models describe capacity as
tations in the focus of attention; see reviews limited by the number of items that can be
by Postle (2006), Ranganath & D’Esposito activated in LTM, which can be thought of as
(2005), Reuter-Lorenz & Jonides (2007), and the bandwidth of attention. However, these
Ruchkin et al. (2003)]. Modeling work and le- models differ on what that number or band-
sion data provide further support for the idea width might be. Cowan (2000) suggested a
that the representations used in both STM limit of approximately four items, based on
and LTM are stored in those regions of cor- performance discontinuities such as errorless
tex that are involved in initial perception and performance in immediate recall when the
encoding, and that frontal activations reflect number of items is less than four, and sharp
processes involved in selecting this informa- increases in errors for larger numbers. (By this
tion for the focus of attention and keeping it view, the classic “seven plus or minus two” is
there (Damasio 1989, McClelland et al. 1995). an overestimate because it is based on stud-
The principle of posterior storage also ies that allowed participants to engage in pro-
allows some degree of reconciliation between cesses of rehearsal and chunking, and reflected
multi- and unitary-store models. Posterior contributions of both the focus and LTM; see


also Waugh & Norman 1965.) At the other of this activity was strongly correlated with es-
extreme are experimental paradigms suggest- timates of each subject’s memory capacity and
ing that the focus of attention consists of a was less pronounced on incorrect than cor-
single item (Garavan 1998, McElree 2001, rect trials, indicating that it was causally re-
Verhaeghen & Basak 2007). We briefly con- lated to performance. Subsequent functional
sider some of the central issues behind current magnetic resonance imaging (fMRI) studies
controversies concerning capacity estimates. have observed similar load- and accuracy-
dependent activations, especially in intrapari-
Behavioral and neural evidence for the etal and intraoccipital sulci (Todd & Marois
magic number 4. Cowan (2000) has re- 2004, 2005). These regions have been impli-
viewed an impressive array of studies leading cated by others (e.g., Yantis & Serences 2003)
to his conclusion that the capacity limit is four in the control of attentional allocation, so it

items, plus or minus one (see his Table 1). seems plausible that one rate-limiting step in
Early behavioral evidence came from stud- STM capacity has to do with the allocation of
ies showing sharp drop-offs in performance attention (Cowan 2000; McElree 1998, 2001;

at three or four items on short-term retrieval Oberauer 2002).
tasks (e.g., Sperling 1960). These experiments
were vulnerable to the criticism that this limit Evidence for more severe limits on focus
might reflect output interference occurring capacity. Another set of researchers agree
during retrieval rather than an actual limit on there is a fixed capacity, but by measuring a
capacity. However, additional evidence comes combination of response time and accuracy,
from change-detection and other tasks that they contend that the focus of attention is
do not require the serial recall of individual limited to just one item (e.g., Garavan 1998,
items. For example, Luck & Vogel (1997) pre- McElree 2001, Verhaeghen & Basak 2007).
sented subjects with 1 to 12 colored squares For example, Garavan (1998) required sub-
in an array. After a blank interval of nearly jects to keep two running counts in STM, one
a second, another array of squares was pre- for triangles and one for squares—as shape
sented, in which one square may have changed stimuli appeared one after another in random
color. Subjects were to respond whether the order. Subjects controlled their own presen-
arrays were identical. These experiments and tation rate, which allowed Garavan to mea-
others that avoid the confound of output- sure the time spent processing each figure
interference (e.g., Pashler 1988) likewise have before moving on. He found that responses
yielded capacity estimates of approximately to a figure of one category (e.g., a triangle)
four items. that followed a figure from the other cate-
Electrophysiological and neuroimaging gory (e.g., a square) were fully 500 millisec-
studies also support the idea of a four-item onds longer than responses to the second of
capacity limit. The first such report was by two figures from the same category (e.g., a
Vogel & Machizawa (2004), who recorded triangle followed by another triangle). These
event-related potentials (ERPs) from subjects findings suggested that attention can be fo-
as they performed a visual change-detection cused on only one internal counter in STM at
task. ERP recording shortly after the onset a time. Switching attention from one counter
of the retention interval in this task indicated to another incurred a substantial cost in time.
a negative-going wave over parietal and oc- Using a speed-accuracy tradeoff procedure,
cipital sites that persisted for the duration of McElree (1998) came to the same conclusion
the retention interval and was sensitive to the that the focus of attention contained just one
number of items held in memory. Importantly, item. He found that the retrieval speed for the
this signal plateaued when array size reached last item in a list was substantially faster than
between three and four items. The amplitude for any other item in the list, and that other

200 Jonides et al.

items were retrieved at comparable rates to of flexibility appears to alter the amount of in-
each other even though the accuracy of re- formation that can be compacted into a single
trieval for these other items varied. representation rather than the total number
Oberauer (2002) suggested a compromise of representations that can be held in STM
solution to the “one versus four” debate. In (Miller 1956). The data of Verhaegen et al.
his model, up to four items can be directly ac- (2004; see Figure 5 of that paper) suggest that
cessible, but only one of these items can be in the latter number still approximates four, con-
the focus of attention. This model is similar to sistent with Cowan’s claims.
that of Cowan (2000), but adds the assump- Building on these findings, we suggest a
tion that an important method of accessing new view of capacity. The fundamental idea
short-term memories is to focus attention on that attention can be allocated to one piece
one item, depending on task demands. Thus, of information in memory is correct, but the

in tasks that serially demand attention on sev- definition of what that one piece is needs to
eral items (such as those of Garavan 1998 or be clarified. It cannot be that just one item is
McElree 2001), the mechanism that accom- in the focus of attention because if that were

plishes this involves changes in the focus of so, hardly any computation would be possible.
attention among temporarily activated repre- How could one add 3+4, for example, if at any
sentations in LTM. one time, attention could be allocated only
to the “3” or the “4” or the “+” operation?
Alternatives to capacity limits based on We propose that attention focuses on what is
number of items. Attempting to answer the bound together into a single “functional con-
question of how many items may be held in text,” whether that context is defined by time,
the focus implicitly assumes that items are the space, some other stimulus characteristic such
appropriate unit for expressing capacity lim- as semantic or visual similarity or momentary
its. Some reject this basic assumption. For ex- task relevance. By this account, attention can
ample, Wilken & Ma (2004) demonstrated be placed on the whole problem “3+4,” allow-
that a signal-detection account of STM, in ing relevant computations to be made. Com-
which STM capacity is primarily constrained plexity comes into play by limiting the number
by noise, better fit behavioral data than an of subcomponents that can be bound into one
item-based fixed-capacity model. Recent data functional context.
from change-detection tasks suggest that ob-
ject complexity (Eng et al. 2005) and similar-
ity (Awh et al. 2007) play an important role in Summary
determining capacity. Xu & Chun (2006) of- What are we to conclude from the data con-
fer neuroimaging evidence that may reconcile cerning the structure of STM? We favor the
the item-based and complexity accounts: In a implication that the representational bases for
change-detection task, they found that activa- perception, STM, and LTM are identical.
tion of inferior intra-parietal sulcus tracked a That is, the same neural representations ini-
capacity limit of four, but nearby regions were tially activated during the encoding of a piece
sensitive to the complexity of the memoranda, of information show sustained activation dur-
as were the behavioral results. ing STM (or retrieval from LTM into STM;
Other researchers disagree with fixed Wheeler et al. 2000) and are the repository
item-based limits because they have demon- of long-term representations. Because regions
strated that the limit is mutable. Practice may of neocortex represent different sorts of infor-
improve subjects’ ability to use processes such mation (e.g., verbal, spatial), it is reasonable to
as chunking to allow greater functional capac- expect that STM will have an organization by
ities (McElree 1998, Verhaeghen et al. 2004; type of material as well. Functionally, mem-
but see Oberauer 2006). However, this type ory in the short term seems to consist of items


in the focus of attention along with recently age, then understanding processing related to
attended representations in LTM. These this limited focus amounts to understanding
items in the focus of attention number no three basic types of cognitive events2 : (a) en-
more than four, and they may be limited to just coding processes that govern the transforma-
a single representation (consisting of items tion from perceptual representations into the
bound within a functional context). cognitive/attentional focus, (b) maintenance
We turn below to processes that operate processes that keep information in the fo-
on these representations. cus (and protect it from interference or de-
cay), and (c) retrieval processes that bring
WHAT PROCESSES OPERATE information from the past back into the cog-
ON THE STORED nitive focus (possibly reactivating perceptual
INFORMATION? representations).

Theoretical debate about the nature of STM
Encoding of items into the focus. Encod-
has been dominated by discussion of structure
ing processes are the traditional domain of

and capacity, but the issue of process is also
theories of perception and are not treated ex-
important. Verbal rehearsal is perhaps most
plicitly in any of the current major accounts
intuitively associated with STM and plays a
of STM. Here we outline three implicit as-
key role in the classic model (Baddeley 1986).
sumptions about encoding processes made in
However, as we discuss below, rehearsal most
most accounts of STM, and we assess their
likely reﬂects a complex strategy rather than a
empirical and theoretical support.
primitive STM process. Modern approaches
First, the cognitive focus is assumed to
offer a large set of candidate processes, includ-
have immediate access to perceptual pro-
ing encoding and maintenance (Ranganath
cessing—that is, the focus may include con-
et al. 2004), attention shifts (Cowan 2000),
tents from the immediate present as well as
spatial rehearsal (Awh & Jonides 2001), up-
contents retrieved from the immediate past.
dating (Oberauer 2005), overwriting (Neath
In Cowan’s (2000) review of evidence in fa-
& Nairne 1995), cue-based parallel retrieval
vor of the number four in capacity estimates,
(McElree 2001), and interference-resolution
several of the experimental paradigms involve
( Jonides & Nee 2006).
focused representations of objects in the im-
Rather than navigating this complex and
mediate perceptual present or objects pre-
growing list, we take as our cornerstone the
sented less than a second ago. These in-
concept of a limited focus of attention. The
clude visual tracking experiments (Pylyshyn
central point of agreement for the unitary-
et al. 1994), enumeration (Trick & Pylyshyn
store models discussed above is that there is
1993), and whole report of spatial arrays and
a distinguishable focus of attention in which
spatiotemporal arrays (Darwin et al. 1972,
representations are directly accessible and
Sperling 1960). Similarly, in McElree’s (2006)
available for cognitive action. Therefore, it is
and Garavan’s (1998) experiments, each in-
critical that all models must identify the pro-
coming item in the stream of material (words
cesses that govern the transition of memory
or letters or objects) is assumed to be repre-
representations into and out of this focused
sented momentarily in the focus.
state.

The Three Core Processes of 2
This carving up of STM processes is also consistent
Short-Term Memory: Encoding, with recent approaches to individual differences in work-
Maintenance, and Retrieval ing memory, which characterize individual variation not in
terms of variation in buffer capacity, but rather in variation
If one adopts the view that a limited focus of in maintenance and retrieval processes (Unsworth & Engle
attention is a key feature of short-term stor- 2007).

202 Jonides et al.

Second, all of the current theories as- preventing aspects of the perceptual present
sume that perceptual encoding into the fo- from automatically entering into the focused
cus of attention results in a displacement of state. Postle (2006) recently found that in-
other items from the focus. For example, in creased activity in dorsolateral prefrontal cor-
McElree’s single-item focus model, each in- tex during the presentation of distraction dur-
coming item not only has its turn in the focus, ing a retention interval was accompanied by a
but it also replaces the previous item. On the selective decrease in inferior temporal cortical
one hand, the work reviewed above regarding activity. This pattern suggests that prefrontal
performance discontinuities after the putative regions selectively modulated posterior per-
limit of STM capacity has been reached ap- ceptual areas to prevent incoming sensory in-
pears to support the idea of whole-item dis- put from disrupting the trace of the task-
placement. On the other hand, as also de- relevant memorandum.

scribed above, this limit may be susceptible In summary, current approaches to STM
to factors such as practice and stimulus com- have an obligation to account for how con-
plexity. An alternative to whole-item displace- trolled processes bring relevant aspects of per-

ment as the basis for interference is a graded ception into cognitive focus and leave others
similarity-based interference, in which new out. It is by no means certain that existing
items entering the focus may partially over- STM models and existing models of percep-
write features of the old items or compete tual attention are entirely compatible on this
with old items to include those featural com- issue, and this is a matter of continued lively
ponents in their representations as a function debate (Milner 2001, Schubert & Frensch
of their similarity. At some level, graded inter- 2001, Woodman et al. 2001).
ference is clearly at work in STM, as Nairne
(2002) and others have demonstrated (we re- Maintenance of items in the focus. Once
view this evidence in more detail below). But an item is in the focus of attention, what
the issue at hand is whether the focus is sub- keeps it there? If the item is in the percep-
ject to such graded interference, and if such tual present, the answer is clear: attention-
interference is the process by which encoding modulated, perceptual encoding. The more
(or retrieving) items into the focus displaces pressing question is: What keeps something
prior items. Although there does not appear in the cognitive focus when it is not currently
to be evidence that bears directly on this is- perceived? For many neuroscientists, this is
sue (the required experiments would involve the central question of STM—how informa-
manipulations of similarity in just the kinds of tion is held in mind for the purpose of fu-
paradigms that Cowan, McElree, Oberauer, ture action after the perceptual input is gone.
and others have used to provide evidence for There is now considerable evidence from pri-
the limited focus), the performance disconti- mate models and from imaging studies on hu-
nuities strongly suggest that something like mans for a process of active maintenance that
displacement is at work. keeps representations alive and protects them
Third, all of the accounts assume that per- from irrelevant incoming stimuli or intruding
ceptual encoding does not have obligatory ac- thoughts (e.g., Postle 2006).
cess to the focus. Instead, encoding into the We argue that this process of mainte-
focus is modulated by attention. This follows nance is not the same as rehearsal. Indeed,
rather directly from the assumptions about the the number of items that can be maintained
severe limits on focus capacity: There must without rehearsal forms the basis of Cowan’s
be some controlled way of directing which as- (2000) model. Under this view, rehearsal is
pects of the perceptual present, as well as the not a basic process but rather is a strategy
cognitive past, enter into the focused state. for accomplishing the functional demands for
Stated negatively, there must be some way of sustaining memories in the short term—a


Rev

Recomendados

Recomendados

Mais conteúdo relacionado

Destaque

Destaque (15)

Semelhante a Rev

Semelhante a Rev (12)

Último

Último (20)

Rev