Oakes_Ch_03.qxd

Continuity of Format and Computation in

Short-Term Memory Development

LISA FEIGENSON

In their first year of life, infants face the enormous task of making sense of

the world around them. Without the ability to store and reason about repre-

sentations of the individual objects, actions, and sounds in their environ-

ment, infants would never accomplish the monumental changes they do.

Storing representations of such individuals in memory allows infants to per-

form computations that, while seemingly simple, are critical to learning

about the world. Comparing a scene to one observed earlier, keeping track of

the presence of objects even when the objects are temporarily occluded, and

making predictions about the outcomes of hidden events are some examples

of such computations. The thesis of this chapter is that the short-term mem-

ory system that enables infants to store object representations, and many of

the computations infants perform over these representations, are continuous

throughout the human life span. Infants and adults show similar capacities

and similar limitations regarding their ability to represent and reason about

objects. At the same time, infants’ and adults’ short-term memory abilities

may differ in some important respects. This chapter explores what is shared

and what may differ in early versus mature short-term memory.

First, I would like to lay out some rough definitions. When I talk about

short-term memory in this chapter, I am referring to the ability to form and

store mental tokens that stand for entities in the outside world. Maintaining

these tokens over short durations allows the entities to be thought about even

when direct perceptual information is absent, as is the case when objects

undergo occlusion. This short-term memory enables infants to represent the

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 51

presence of entities when those entities might be temporarily hidden (e.g.,

“There is an object under that blanket”) and to store information about those

entities (e.g., “The object under the blanket is round and red and striped”).

Furthermore, storing representations of more than one item at a time enables

infants to compute across an entire scene (e.g., “There are three balls under

the blanket”), rather than over just a single item. Here, I suggest that this type

of short-term memory in infants corresponds to a system of short-term mem-

ory that has been studied in adults (for an argument that working memory is

a better term for this same system, see chapter 1, this volume). Indeed, one

of the most striking observations about this memory system is the extent to

which it remains constant throughout development, both in its capacity and

in the computations it supports.

I begin with an exploration of the limits on infants’ and adults’ short-

term memory capacity; the evidence I review builds the case that the very

same system of memory is relied on across the life span. Second, I examine

findings that infants and adults perform similar computations over these

short-term memory representations. Both groups compute the continuous

and discrete properties of object arrays, and both groups use chunking as a

means to recode memory representations into a more efficient format.

Third, I address a possible developmental difference in short-term memory,

asking whether infants and adults differ in the degree to which their mem-

ory computations are driven in top-down versus bottom-up fashion. Finally,

I close the chapter by raising some outstanding questions and by suggesting

some avenues for future research on short-term memory development.

Short-Term Memory for Object Arrays

Some 25 years ago, pioneers in the newly emerging field of infant cognition

demonstrated that, contrary to Piaget’s claims, young infants represent and

reason about objects. Critically, they do so in ways that go beyond the im-

mediate sensory data available to them. By 5 to 7 months of age, infants

represent objects that have been covered by cloths, hidden by screens, or

concealed in darkness (e.g., Baillargeon, Spelke, & Wasserman, 1985; Hood

& Willats, 1986; Shinskey & Munakata, 2003). That infants have stored

representations of these objects in memory is shown by their continued

reaching for the objects once hidden, or by their longer looking when objects

unexpectedly disappear. Even more impressively, infants reason over these

stored representations of objects. For example, when a solid object is placed

behind a screen and a second object is launched on a direct path toward it,

infants look longer when the second object emerges magically unscathed

from the other side of the screen (Baillargeon, 1986). Apparently, infants

have reasoned that one solid object cannot pass through another. Because

infants’ looking times in situations such as these depend on inferred inter-

actions between objects that are hidden rather than visible, we conclude

52 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 52

that infants are operating on representations stored in memory rather than

operating directly on the immediate sensory data.

Researchers have attempted to characterize the memory systems infants

use to reason about such instances of object behavior. One question of inter-

est is whether infants reason over object representations stored in short-term

memory or long-term memory.

Although the distinction between short- and

long-term memory systems has often been controversial, many have suggested

that the amount of information stored by each person over long durations is

too large and unwieldy to allow sufficiently rapid access for the moment-to-

moment comparisons we constantly perform, and which characterize the

occlusion events typical of experiments with infants. This problem motivates

the existence of a system that is distinct from the larger, long-term memory

store. This system holds a limited amount of information in a temporary state

of privileged access. Representations held in this short-term memory system

can be formed quickly but decay over time, whereas long-term memory rep-

resentations take longer to form but are far more enduring (see chapters 7–10,

this volume). On some models, the information held in short-term memory

can come either from the outside world (e.g., storing a representation of an

object that is currently visible), or it can come from the activation of a repre-

sentation previously stored in long-term memory (e.g., thinking about an

object that was seen yesterday; Cowan, 2001). In either case, information held

in short-term memory is available for immediate processing.

Several factors hint that short-term memory does underlie infants’ reasoning

about the kinds of object arrays typical of infant cognition research. In such

studies, infants receive only brief exposure to a scene before objects are hidden

from view, perhaps limiting the extent to which they have the opportunity to

store long-term representations. In addition, experiments manipulating the

delay between an object’s disappearance from view and the moment when in-

fants are allowed to retrieve it reveal that infants’ object memories fade rapidly

(Diamond, 1990). These factors begin to suggest that short-term memory is the

likely locus of infants’ object-tracking abilities. However, most studies investi-

gating infants’ object representations have not been designed to distinguish the

relative contributions of short- versus long-term memory. For example, they

have systematically manipulated neither the duration of infants’ object

exposure nor the interval over which infants must maintain the object repre-

sentations in memory, at least not in ways that would bear decisively on which

memory system is involved. Therefore, the conclusion that many studies of in-

fants’ object representations are tapping short-term memory remains tentative.

Short-Term Memory Capacity in Adults

Further evidence is needed. Measuring the capacity of infants’ memory is a

potential source of such evidence, since capacity differences have tradition-

ally been a distinguishing characteristic of short- versus long-term memory.

While long-term memory is usually thought of as unlimited, short-term

Continuity of Format and Computation 53

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 53

memory is thought to store only a small amount of information at any one

time. This notion of a limited-capacity short-term memory store originated

with James (1890) and received significant attention following Miller’s

(1956) influential postulation of a “magical number seven, plus or minus

two.” Miller made famous the view of short-term memory as a repository

limited by the number of unique items it can hold (approximately seven, in

Miller’s view), rather than by overall informational load (where information

load is influenced by factors such as the complexity of the items). Since

then it has been suggested that “seven, plus or minus two” probably over-

estimates short-term memory capacity and is likely the by-product of addi-

tional mental processes, such as chunking, that allow subjects to recode

individual items into groups. A more reasonable estimate, obtained when

chunking is prevented, is three to four (Cowan, 2001).

Cowan (2001) reviewed a wide range of experiments probing short-term

memory capacity, and finds that most of these produce estimates of three to

four items. Space precludes presenting that evidence here, but a summary

of a classic experimental series by Sperling (1960) illustrates the type of

results obtained. Sperling showed adult subjects 3 ! 4 arrays of letters,

presented too briefly for the subjects to store all 12 letters in long-term mem-

ory. On whole-report trials, subjects reported the names of all of the letters

they could remember; they averaged around 4. On partial-report trials,

subjects reported only a subset of the array as specified by an auditory cue.

When the cue was heard 2 to 5 seconds after the array had disappeared, sub-

jects were able to report an average of 1.3 of the 4 letters in the cued row.

Multiplied by the number of rows in the array (3), this again yielded approx-

imately 4 as the upper limit on short-term memory capacity.

More recently, short-term memory tasks produced a similar capacity limit

in adults. Halberda, Simons, and Wetherhold (2006) showed subjects a rap-

idly flashing grid of 32 dots, each of a different luminance value. All but one

of the dots maintained its individual luminance value from flash to flash; the

remaining dot oscillated between two different values. Subjects had to find

the single changing dot. Halberda et al. found that subjects were able to en-

code the luminance values of a subset of the dots on each flash of the array,

store them in memory, and compare them to the dots’ luminance values on

the next flash. Subjects examined subsets of dots in this way until they

located the changing item. By analyzing the number of flashes required to

locate the target dot, Halberda et al. estimated that the number of dots subjects

could store in short-term memory from a single flash of the array was three.

Other findings strengthen the claim that the capacity limit is defined by the

discrete number of items being held in memory, rather than by total informa-

tion load. Luck and Vogel (1997) used a change detection task in which sub-

jects received a 100-millisecond visual exposure to an array of items, followed

900 milliseconds later by a test array. Subjects had to report whether the two

arrays were identical or whether any of the items had changed their features.

Luck and Vogel found that performance was at ceiling for arrays containing

one to three items, and declined with sets of four or more. Most strikingly,

54 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 54

subjects were just as accurate for arrays of items that contained multiple fea-

tures (color and texture orientation) as for arrays containing just a single fea-

ture (color). Thus, the number of items in the array, and not the number of

features in the array, determined subjects’ memory capacity.

Several researchers have suggested that there is a similar three-to-four-

item capacity limit in attention, prior to the storage of any items in short-

term memory (Carey, 2004; Scholl & Leslie, 1999; Trick & Pylyshyn, 1994).

This view derives largely from results of the multiple object tracking para-

digm. This paradigm was developed by Pylyshyn to examine the process by

which a subset of the information in a scene achieves priority for further

processing, before its transfer into memory (Pylyshyn, 1994, 2001). In the

multiple object tracking task, subjects track several moving onscreen targets

amid a field of identical distractors. Because no featural cues distinguish the

targets from the distractors, the only way for subjects to succeed is to attend

to the targets from the start of each trial (when targets flash briefly to

indicate their status) and to keep attending to them in parallel as they move

haphazardly through the scene. Subjects perform this task effortlessly with

one, two, three, and often four targets. But when asked to track more than

four, performance plummets (Pylyshyn & Storm, 1988). Because this task

was designed to require little or no memory, many have concluded that the

observed limit is grounded in attention. However, more recent evidence dis-

putes the view of a three-to-four-item limit on attention. Experiments by

Alvarez and Franconeri (2005) suggest that attentional capacity increases to

well above four items when the items move more slowly. Thus it remains to

be seen whether an item-based limit on attention will hold.

If such a limit does hold up in purely attentional tasks, is the existence of an

identical three-to-four-item limit that constrains both attention and short-term

memory purely coincidental? An alternative view has been offered by Cowan

(2001), who suggests that there is no structural distinction between attention

and short-term memory. Instead, Cowan suggests that in order to reason about

remembered items, the items need to be pulled from memory storage into the

“focus of attention.” This focus of attention can be thought of as activation of

the stored items, where activation is required for any sort of conscious process-

ing. Cowan suggests that while memory storage itself is unlimited, only three to

four items can be brought into the focus of attention at any given time. Cowan’s

proposal is controversial but serves to highlight the difficulty in distinguishing

between capacity limits on attention versus those on short-term memory.

Measuring Short-Term Memory in Infancy:

Recent Advances

We now return to the question of which memory system underlies infants’

ability to track and reason about hidden objects. Given that adults can main-

tain three to four items in short-term memory, a similar limit on infants’

abilities would be important in illustrating continuity across development.

Recent findings have obtained just such a limit. In a modified version of

Continuity of Format and Computation 55

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 55

Luck and Vogel’s (1997) change-detection paradigm, Ross-Sheehy, Oakes,

and Luck (2003) presented 10-month-old infants with cycles of simple ob-

ject arrays appearing simultaneously on a pair of adjacent screens. On every

cycle, the arrays appeared on each screen for 500 milliseconds, disappeared

for 250 milliseconds, then reappeared for 500 milliseconds. On one of the

two screens, one of the objects in the array changed color during the 250-

millisecond retention period. On the other screen, none of the objects

changed. The arrays cycled such that one screen always displayed an array

that changed during the retention period, while the other screen always dis-

played an array that remained constant (for a more detailed description of

their experimental design, see chapter 4, this volume).

Ross-Sheehy et al. (2003) reasoned that since infants generally prefer to look

at more complex displays rather than at simple ones, they would spend more

time looking at the screen with the changing array than at the screen with the

constant array. The ability to notice a change in the array from cycle to cycle

depended on storing a representation of the objects in the array, maintaining

this representation over the 250 milliseconds when no display was visible, and

then comparing it to the next array that appeared. Therefore, a preference for

looking at the changing screen implies successful memory retention of the fea-

tures of all of the objects in the array (since which particular object changed

varied randomly from cycle to cycle). Ross-Sheehy and colleagues used this

paradigm to probe memory limits by comparing infants’ performance with

arrays containing different numbers of objects. They found that with one-, two-

, three-, and four-object arrays, infants preferred to look at the changing screen.

But with six-object arrays, infants showed no such preference. Apparently, 10-

month-old infants were unable to represent more than four items at a time and

therefore did not discriminate the changing from the unchanging array.

This work, using methods that closely resemble those used to study

adults’ visual short-term memory, suggests that infants, like adults, can store

representations of three to four items at a time. But what about memory for

the kinds of real, three-dimensional object arrays used in so many experi-

ments on infant cognition, and which constitute the majority of infants’

natural daily experience? Objects in such arrays are likely to be more com-

plex in their shape, shading, and features than the simple squares used by

Ross-Sheehy et al. (2003). In addition, objects in a natural scene often un-

dergo complex patterns of motion, sometimes involving periods of occlusion

during which they might be hidden for several seconds at a time. Will the

three-to-four-item limit of short-term memory also be observed when infants

are faced with naturalistic object arrays?

Infants’ Short-Term Memory for Naturalistic

Object Arrays

The question of exactly how many such hidden objects infants can remem-

ber and reason over was first raised by Karen Wynn, who demonstrated that

56 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 56

by 5 months of age infants already can represent at least two occluded

objects. After seeing a screen hide one object, followed by the addition of a

second object behind the screen, infants expected to see two objects when

the screen was lifted (Wynn, 1992). Work using similar looking time meth-

ods to measure infants’ representation of occluded objects has confirmed

that infants can remember at least two hidden objects at a time (Kaldy &

Leslie, 2003; Koechlin, Dehaene, & Mehler, 1997; Simon, Hespos, & Rochat,

1995; Uller, Huntley-Fenner, Carey, & Klatt, 1999).

My colleagues and I have extended these findings by asking just how

many such hidden objects infants can remember. We probed the upper lim-

its of infants’ ability to track occluded objects by creating a simple task in

which the number of objects infants had to store in memory was parametri-

cally varied. In this “cracker choice” task (Feigenson & Carey, 2005;

Feigenson, Carey, & Hauser, 2002), 10- and 12-month-old infants saw two

quantities of desirable objects (graham crackers) sequentially placed into a

pair of opaque buckets and then were allowed to choose between them.

Since determining which bucket contained more crackers required main-

taining and comparing representations of the hidden objects, and since

adults have been shown to store object representations in short-term mem-

ory for durations comparable to those we used (approximately 8–10 sec-

onds; Noles, Scholl, & Mitroff, 2005), we reasoned that our procedure would

serve as a naturalistic test of preverbal children’s short-term memory.

We started by giving infants a choice between one or two, two or three,

or three or four crackers. Groups of 16 different 10-month-old infants and 16

different 12-month-old infants participated in each of these numerical com-

parisons. Infants’ spontaneous, untrained abilities were revealed by giving

each infant just one opportunity to make a choice; thus, the experiment con-

sisted of a single trial for each participant. In our experimental procedure,

infants sat on the floor across from an experimenter. The experimenter

produced two opaque plastic buckets, showed infants that they were empty,

and placed them on the floor approximately 70 cm from infants’ starting

location and approximately 70 cm from each other. The experimenter then

placed crackers one at a time into the buckets, making sure that infants

attended to the placement of each cracker. For example, in a one-or-two

choice, the experimenter placed one cracker in one bucket and two crackers

one at a time into the other bucket. Which side the presentation began on

and which bucket received the greater number of crackers was counterbal-

anced across participants. The dependent measure was simply which

bucket infants chose to walk or crawl to.

Figure 3.1 displays the pattern of infants’ spontaneous choices. We found

that with choices of one or two and two or three crackers, infants of both age

groups successfully chose the bucket containing the greater quantity. Infants

failed, however, with a choice of three or four crackers. Infants’ failure with

this comparison might have been due to either the less discriminable ratio

between the quantities, or to the quantities having exceeded the maximum

Continuity of Format and Computation 57

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 57

number of items infants could hold in memory. Therefore, we next tested

new groups of infants with quantities in a ratio with which we had already

observed success (1:2), but which involved more than three items in a single

location. We gave infants a choice between two or four, three or six, and one

or four crackers. Since no age differences between 10- and 12-month-olds

were observed in any of the previous comparisons we tested, each of the

comparisons included 16 infants ranging between 10 and 12 months.

Infants failed to choose systematically with any of these quantities (see

Figure 3.1). This dramatic breakdown in performance illustrates that infants’

ability to remember the hidden objects was determined by the total number of

objects seen, and not by ratio of differences between the two quantities we pre-

sented. Infants succeeded only when one, two, or three crackers were placed

in either bucket, and chose entirely by chance when required to remember

larger numbers (Figure 3.1). A variety of control conditions ensured that the

abrupt break in infants’ performance was, in fact, due to the number of objects

presented and not to the total presentation duration or complexity (see

Feigenson, Carey, & Hauser, 2002, for details). Thus, it appears that in this task

infants were limited to storing up to three items in each hiding location.

58 Short-Term or Working Memory

100

Comparison

10 mos:

1 vs. 2

12 mos:

1 vs. 2

10 mos:

2 vs. 3

12 mos:

2 vs. 3

10 mos:

3 vs. 4

12 mos:

3 vs. 4

2 vs. 4

3 vs. 6

1 vs. 4

% Choosing Greater Quantity

FIGURE 3.1. The percentage of infants choosing the greater of two

quantities. Infants chose the greater quantity with small arrays, but failed

whenever either array contained four or more objects. Reprinted from

Feigenson, L., Carey, S., & Hauser, M. (2002). The representations

underlying infants’ choice of more: Object files versus analog magnitudes.

Psychological Science, 13(2), 150–156, copyright (2002), with permission

from Blackwell; and from Feigenson, L., & Carey, S. (2005). On the limits

of infants’ quantification of small object arrays. Cognition, 97, 295–313,

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 58

This naturalistic cracker choice task, which required infants to track real,

three-dimensional objects undergoing motion and occlusion, revealed the

same three-item limit on performance as has been observed in tasks show-

ing infants (Ross-Sheehy et al., 2003) and adults (Halberda et al., 2006;

Halberda, Sires, & Feigenson, 2006; Luck & Vogel, 1997) simple, computer-

ized arrays. This suggests that the three-to-four-item capacity limit applies

to a range of entities, from grayscale dots to real moving objects. However,

one difference between our cracker choice task and previous tasks assessing

short-term memory lies in the timing of the presentation. Our cracker task

involved sequentially presented objects, whereas previous tasks presented

infants and adults with objects that were all visible at once. Therefore, our

next step was to ask whether the three-to-four-item capacity limit would be

found when real, three-dimensional objects are simultaneously presented.

We addressed this question by again measuring the number of hidden

objects infants could remember, but with a simultaneous rather than sequen-

tial presentation. In our manual search paradigm, infants searched for objects

they had seen an experimenter hide in an opaque box (Feigenson & Carey,

2003, 2005). A group of 12- and 14-month-old infants saw one to four iden-

tical balls simultaneously visible atop the box; the balls were then picked up

and inserted through a cloth-covered opening in the box’s front face. After-

ward, infants were allowed to reach in and retrieve the balls. Unbeknownst

to the infants, on some trials the experimenter surreptitiously removed a sub-

set of the balls from a concealed opening in the back of the box. We meas-

ured infants’ continued searching and compared it to their baseline level of

searching on trials when the box was expected to be empty. Any increased

searching suggests that infants successfully represented and were searching

for the remaining object or objects inside the box. In this way, our manual

search task serves as a measure of the number of occluded items infants can

remember over a relatively short duration.

We probed the limit on the number of objects infants could simultane-

ously remember via a series of x versus y comparisons. For any x versus y

comparison, infants’ searching after they saw x balls hidden and had retrieved

x of them was contrasted with their searching after they saw y balls hidden

and had retrieved only x of them. The logic can be illustrated with a one-

versus-two comparison. On one-object trials, infants saw the experimenter

hide a single ball in the box, were then allowed to retrieve it, and any subse-

quent searching into the now-empty box was recorded during the 10 second

measurement period that followed (Figure 3.2a). This was compared to the

duration of searching on two-object trials, on which infants saw two identi-

cal balls hidden and then were allowed to retrieve just one of them. While

the experimenter surreptitiously held the remaining ball out of reach for 10

seconds, any searching for the “missing” ball was recorded. After 10 seconds,

the experimenter retrieved the remaining ball and showed it to infants, after

which the box was once again empty. Any further searching was recorded in

a final 10 second measurement period (Figure 3.2b).

Continuity of Format and Computation 59

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 59

1) Box is placed on table.

2) Experimenter places 2 balls on box,

then hides them inside.

3) Infant allowed to retrieve 1 ball.

Experimenter surrepitiously removes 2nd ball.

4) 2-Object (1 r emaining) trial: Infant s

searching is measured. 1 ball expected inside.

5) Experimenter finds 2nd ball.

6) 2-Object (expected empty) trial: Infant s

searching is measured. Box expected empty

1) Box is placed on table.

2) Experimenter places 1 ball on box,

then hides it inside.

3) Infant allowed to retrieve 1 ball.

4) 1-Object (expected empty) trial: Infant s

searching is measured. Box expected empty

FIGURE 3.2. Presentation sequences illustrating: (a) a one-object trial; (b) a two-object trial. Reprinted from Feigenson, L., &

Carey, S. (2003). Tracking individuals via object-files: Evidence from infants’ manual search. Developmental Science, 6,

568–584, copyright (2003), with permission from Blackwell; and from Feigenson, L., & Carey, S. (2005). On the limits of infants’

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 60

If infants were able to remember the correct number of objects hidden in

the box, they should search the box only when the box was expected to still

contain one or more objects. Therefore, we assessed infants’ performance by

examining the difference in their searching on trials when the box was ex-

pected to contain more objects versus trials when the box was expected to be

empty. For example, subtracting search time after infants had seen one object

hidden and had retrieved it from search time after infants had seen two objects

hidden and had retrieved just one of them creates a difference score. If infants

represent two as more than one, this difference score should be positive. We

found that when 12- and 14-month-old infants were presented with this task,

they succeeded (i.e., had positive difference scores) with one-versus-two and

two-versus-three comparisons

, but failed with two-versus-four and one-

versus-four comparisons (Feigenson & Carey, 2003, 2005). When infants fail,

we observe difference scores that are not different from chance (Figure 3.3).

Taken together, the results from the experiments just reviewed using change

detection (Ross-Sheehy et al., 2003), cracker choice (Feigenson & Carey, 2005;

Feigenson, Carey & Hauser, 2002), and manual search (Feigenson & Carey, 2003,

2005) have yielded identical patterns of results concerning infants’ capacity

limits. Whether infants saw food or nonfood objects, two-dimensional or three-

dimensional objects, sequential or simultaneous presentation, or were asked to

approach the larger of two total quantities, to search for hidden objects, or to

Continuity of Format and Computation 61

FIGURE 3.3. Difference scores (searching when the box contained more objects

minus searching when the box was empty) reflect 12- to 14-month-old

infants’ capacity to represent and discriminate arrays containing different

numbers of objects (Feigenson & Carey, 2003, 2005).

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 61

notice changing features in an object array, infants were limited to representing

approximately three to four objects at a time in memory.

An important note concerning these observations of short-term memory

capacity is that the three-to-four-item limit emerges in situations that

require the tracking of objects qua individuals. In contrast, other tasks re-

quire infants to represent an array of items as a group whose members are

not stored individually, as with a set of 16 dots scattered randomly on a

screen. In such tasks, infants have been shown to represent the approximate

numerosity of the whole array (Lipton & Spelke, 2003; McCrink & Wynn,

2004; Xu, 2003; Xu & Spelke, 2000; Xu, Spelke, & Goddard, 2005), but likely

are not representing the individual dots that comprise the array. It appears

that the ability to represent the approximate numerosity of a set containing

many items and the ability to represent a small number of discrete individuals

are subserved by different mental processes.

Several aspects of infants’ performance with small versus large arrays sup-

port the view that infants represent them in fundamentally different ways.

First, with large numerosities, such as Xu and Spelke’s arrays of 16 or 24 dots,

infants’ success or failure depends on the ratio between to-be-discriminated

arrays, rather than on the absolute number of items presented. For example,

6-month-old infants discriminate arrays of 16 from arrays of 32, but not from

arrays of 24. We have already seen that for small arrays the reverse is true: It

is the absolute number of items that determines performance (Feigenson &

Carey, 2003, 2005; Feigenson, Carey, & Hauser, 2002). Second, infants often

appear unable to represent the approximate numerosity of arrays containing

four or fewer objects when the arrays are controlled for continuous properties

that frequently correlate with number (Feigenson, Carey, & Hauser, 2002; Xu,

2003; Xu et al., 2005). In contrast, infants can represent the approximate

numerosity of arrays that are controlled for continuous properties when the

arrays contain large numbers of items (Lipton & Spelke, 2003; McCrink &

Wynn, 2004; Xu, 2003; Xu & Spelke, 2000; Xu et al., 2005). These two diver-

gent patterns of results suggest that two distinct mental systems are available

to infants. One of these systems enables the representation of the approximate

numerosity of large arrays. The other, which is the focus of the present dis-

cussion, allows precise representations of one to four items to be held in

short-term memory (for further discussion of this two-system view, see

Feigenson, Dehaene, & Spelke, 2004).

Further Evidence of Continuity: Increasing

Short-Term Memory Storage via Chunking

Chunking in Adults

Besides claiming that adults have a limited-capacity short-term memory,

Miller (1956) also suggested that this capacity could sometimes be increased

by condensing information into a more efficient format. Specifically, Miller

62 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 62

proposed that individual items could be bound together in memory into

chunks whose components were in some way related to each other. If these

chunks, rather than the individual items that comprised them, occupied the

limited number of available memory slots, then the chunks could later be

“unpacked” into their constituent components. Hence, the overall amount of

information accessible via short-term memory could effectively be increased.

That chunking can indeed increase memory capacity in this way has been

shown in impressive demonstrations of memory enhancement. One particu-

lar subject, S.F., was able to increase his memory span to nearly 80 random

digits (Ericcson, Chase, & Faloon, 1980). S.F. had an entirely normal memory

span at the start of the experiment, but after over 200 hours of laboratory

practice he had a span equal to that of professional memory experts. S.F.

accomplished this dramatic improvement by developing and perfecting his

own idiosyncratic chunking strategy. He associated every three or four digits

presented to him with a meaningful unit of information already present in

his long-term memory. For example, S.F. remembered the digits 3, 4, 9, and

2 as “3 minutes and 49 point 2 seconds,” which he knew was a near-record

time to run the mile. The digits 1, 9, 4, and 4 were recalled as “1944, near the

end of World War II.” Interestingly, even by chunking four-digit strings into

discrete chunks, the three-to-four-item limit on short-term memory should

have prohibited S.F. from storing any more than three-to-four chunks con-

taining a total of 12 to 16 individual digits. S.F. surpassed this expected limit

by creating hierarchical memory entries in which chunks were nested within

“superchunks.” This extremely efficient collapsing of information accounted

for S.F.’s impressive memory abilities. S.F.’s short-term memory enhance-

ment is not an isolated case. Indeed, the finding that adults can use semantic

information to increase storage has also been obtained with naive subjects.

Typical college students were able to increase their short-term memory for

digits over several laboratory sessions by associating groups of digits with

preexisting referents, just as S.F. did (Chase & Ericsson, 1981).

A similar chunking mechanism has been found to underlie the excep-

tional performance of chess experts, who show vastly better memory for the

configuration of pieces on a chessboard than do nonexperts. Rather than

having a greater number of memory slots in which to store the individual

pieces’ locations, these experts benefited from the ability to chunk multiple

pieces into recognizable formations (Simon & Chase, 1973). Doing so

allowed them to store the entire board in terms of only a few formations, the

individual components of which could be reconstructed from long-term

memory. Support for the explanation that semantic knowledge allowed the

formation of chunks comes from experiments testing the memory of expert

versus novice players for randomly positioned pieces, as opposed to mem-

ory for configurations that might occur in an actual chess game. When pre-

sented with random configurations, the recall performance of experts was

no better than that of novices (Simon & Chase, 1973).

Thus, the evidence indicates that adults can use existing knowledge to

condense individual bits of information into larger chunks. Doing so enables

Continuity of Format and Computation 63

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 63

the storage of more total units of information (e.g., more random digits or

more locations on a chessboard). This is because short-term memory only

has to retain the chunks themselves (e.g., a near-record time for running the

mile), since the individual components of these chunks (e.g., 3, 4, 9, 2)

already exist in long-term memory. Such a computation is clearly useful,

allowing greater speed and efficiency of access to information across a wide

variety of situations. But what are the origins of chunking? Is this highly use-

ful strategy a learned one, perhaps acquired during formal education? Or is

it available early on, prior to explicit instruction?

Chunking in Infants

My laboratory has addressed this question in a series of studies examining

chunking in 14-month-old infants. Infants of this age had previously

demonstrated a three-object memory limit in the manual search task, as dis-

cussed earlier (Feigenson & Carey, 2003, 2005). The new question was

whether infants in this task could be induced to chunk individual items into

smaller sets and thereby increase the total number of items remembered. To

test this, we presented infants with arrays of identical objects that we then

hid inside a box (Feigenson & Halberda, 2004). We used the same one-ver-

sus-two and two-versus-four object comparisons as in previous experiments

(Feigenson & Carey, 2003, 2005). On one-versus-two comparisons, we asked

whether infants searched the box more after seeing two objects hidden and

retrieving just one of them (the other was surreptitiously withheld) than

they did after seeing one object hidden and retrieving one. Success would

indicate the ability to remember at least two objects, and to recognize that

two is more than one. On two-versus-four comparisons, we asked whether

infants searched the box more after seeing four objects hidden and retriev-

ing just two of them (the other two were surreptitiously withheld) than they

did after seeing two objects hidden and retrieving two. Here, success would

indicate the ability to remember up to four objects, and the recognition that

four is more than two.

Earlier in this chapter, I explained that infants had previously failed at

this kind of two-versus-four comparison when four objects were presented

in a single line on top of the box. The new manipulation in this study was

the spatial arrangement of the objects prior to hiding (see Figure 3.4). On

some trials, all of the objects were presented in a single location centered on

top of the box (e.g., four objects in a line atop the box). On other trials, the

objects were presented on two spatially separated platforms located on

either side of the box (e.g., two objects on the left-hand platform, and two

on the right-hand platform). Our hypothesis was that this spatial grouping

cue would help infants chunk four objects into two sets of two, thereby en-

abling them to successfully represent a total of four items at once.

We found that this spatial grouping changed the total number of objects

infants were able to remember. Although infants succeed at distinguishing

64 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 64

FIGURE 3.4. Difference scores (searching when the box contained more objects minus searching when the box was empty).

Difference scores significantly above 0 reflect 14-month-old infants’ capacity to remember the correct number of objects in the

box. In (a), infants successfully remembered two objects in the box but failed to remember four. In (b), objects were spatially

separated into two sets, each containing fewer than three objects. Infants again remembered two objects, and also successfully

remembered four objects. In (c), infants’ failure to remember four objects and their success at remembering two sets of two was

replicated within-subject. In (d), infants remembered the precise location of each set when the sets were hidden in spatially

separate locations. Reprinted from Feigenson, L., & Halberda, J. (2004). Infants chunk object arrays into sets of individuals.

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 65

one versus two no matter how the objects were arranged, they overcame the

three-item short-term memory limit only on trials when the objects were

spatially grouped. Only when objects were presented in two distinct groups

of two did infants distinguish the hiding of two objects from the hiding of

four (Figure 3.4). This pattern reveals two things. First, it replicates our pre-

viously reported three-item limit on infants’ tracking of occluded objects

(Feigenson & Carey, 2003, 2005), showing that infants were unable to store

a single four-object array in memory. Second, these results also show that

this limit can sometimes be overcome. By chunking representations of indi-

vidual items into smaller units, infants were able to remember more total

objects. The 14-month-old infants represented two chunks, each containing

two individual objects. To our knowledge, this is the first demonstration of

chunking in infants.

Top-Down Versus Bottom-Up Computations

Initiating Chunking

The data reviewed above suggest that the chunking operations that were clas-

sically studied by Miller and others may be both independent of formal train-

ing and available quite early in life. If so, this clearly strengthens the case for

the continuity of short-term memory throughout the life span, as both the lim-

its on short-term storage and the chunking used to overcome these limits

appear to be present in infants as well as in adults. However, there is an im-

portant difference between the finding that adults can increase short-term

memory capacity and the finding that infants can do so. The adult chunking

studies show that adults can use semantic information to condense informa-

tion. For example, adults can bind multiple individual items together based on

existing knowledge (as with the race time example) or can recognize multiple

individuals as forming a meaningful gestalt (as with the chess expert example).

These computations rely on semantic knowledge that is available for associa-

tion with the objects in the array. The computations thus can be considered

top-down, in the sense that they are driven from the internal knowledge to the

external, to-be-chunked items in the world.

Can infants use semantic knowledge to drive chunking? Our developmen-

tal results (Feigenson & Halberda, 2004) show that the spatial organization of

an array into sets, each of which contains three or fewer items, helps infants

overcome the three-to-four-item limit on short-term memory. However, un-

like S.F. or the chess experts, infants relied on spatiotemporal rather than

semantic information. Furthermore, the computation they performed was bot-

tom-up in the sense that the requisite information for chunking was present

in the array itself, rather than in infants’ existing knowledge. The spatial

arrangement that the experimenter imposed on the array led infants to parse

it into smaller sets, rather than infants themselves imposing their knowledge

66 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 66

to reorganize their representation of the array. Thus far, it is an open question

whether infants can also chunk in a top-down fashion on the basis of stored

semantic knowledge.

Experiments in my lab are currently exploring which sources of informa-

tion infants can use to chunk an array. As in our earlier investigations, we

take as evidence for chunking the ability to represent a total of four objects

in our manual search task. These previous studies revealed that infants could

do so only if they saw the four-object array presented as two spatially sepa-

rated sets, each containing fewer than three items (Feigenson & Halberda,

2004). In a series of new studies, we replace spatial information with seman-

tic information as the potential basis for chunking. Infants see all four objects

in a single line atop the box—a spatial arrangement that has previously led in-

fants to fail. However, we now show infants an array of two cars and two cats,

instead of the four identical balls we used in our earlier studies (Feigenson &

Carey, 2003, 2005; Feigenson & Halberda, 2004). Given that 14-month-old in-

fants are reported by their parents to be familiar with these entities (and given

that most 14-month-olds already know the words car and cat, or kitty; Fenson

et al., 1994), we hypothesize that infants may be able to use this semantic

knowledge to chunk the four-object array into two sets of two. To ask whether

any observed success is based on semantic knowledge of the object categories,

as opposed to low-level perceptual differences between the two types of

objects, on other trials we present infants with two sets of two objects that are

perceptually distinct, yet from unfamiliar categories. If infants fail to repre-

sent all four objects when presented with unfamiliar objects, such as two toy

shrimp and two toy tanks, but succeed with two cars and two cats, then we

can more confidently say that infants are able to use semantic knowledge in a

top-down fashion to chunk the array.

Initiating Computations of Discrete and

Continuous Quantity

The question of whether infants can initiate top-down computations over

object representations is not exclusive to chunking, but also arises for other

operations performed over representations being held in memory. An exam-

ple comes from the work on infants’ computations of quantity. Infants have

been shown to be capable of computing the discrete number of individual

objects in object arrays, showing different looking patterns to expected versus

unexpected numbers of objects (Cheries, DeCoste, & Wynn, 2003) or search-

ing a box until the expected number of objects has been retrieved (Feigenson

& Carey, 2003). These findings obtain when the total area or volume of the

arrays is controlled for. Infants also can compute the total continuous extent

contained in an array, showing increased looking when the overall summed

area or perimeter of the items in the array changes (Clearfield & Mix, 1999,

2001; Feigenson, Carey, & Hauser, 2002) or choosing to approach an array

containing a greater total volume of food over an array containing a smaller

Continuity of Format and Computation 67

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 67

total volume, regardless of the number of individual objects involved

(Feigenson, Carey, & Spelke, 2002). That these quantity computations oper-

ate over the same short-term memory representations discussed above is

shown by the conditions under which infants successfully perform them.

Infants can compute discrete or continuous quantity over arrays containing

small numbers of objects. But when arrays contain four or more objects, in-

fants fail to compute either number or total extent (Feigenson & Carey, 2003,

2005; Feigenson, Carey, & Hauser, 2002; Xu, 2003; Xu et al., 2005). Thus, rec-

ognizing the number of objects in an array and recognizing the total extent

contained in the array are both the output of computations performed over

short-term memory representations of objects. When there are too many ob-

jects to be represented in short-term memory, infants fail to compute either

number or total extent.

What prompts infants to respond to the discrete (e.g., number of individ-

ual objects) versus continuous (e.g., color, total extent) properties of a given

object array? Although this question will likely be the focus of many future

experiments, one recent set of findings suggests that the features of the

objects themselves play a role in determining which dimension of quantity

infants represent. A group of 7-month-old infants was habituated to object

arrays, then tested with arrays in which either the number of objects or the

total surface area had changed (Feigenson, 2005; Feigenson, Carey, & Spelke,

2002). The results revealed that when the array contained objects that were

identical in color, pattern, and texture, infants dishabituated to changes in

the total area of the array, but not to changes in the number of objects in the

array. When the array contained objects that contrasted with each other in

color, pattern, and texture, however, infants did just the reverse. They disha-

bituated to changes in the number of objects in the array but not to changes

in total area. In other words, infants appeared able to compute either num-

ber or surface area but unable to perform both computations over the same

array. Figure 3.5 depicts this double dissociation.

In these experiments, the computation that infants performed (number

versus total extent) appeared to be under exogenous control, influenced in

a bottom-up fashion based on whether objects in the array had identical

properties or not. In contrast, adults can exert top-down control over these

computations, choosing whether to represent the number or extent con-

tained in an array even on a trial-by-trial basis (Feigenson & Halberda, in

preparation). Thus, infants and adults may differ in the endogenous versus

exogenous nature of the quantity computations they can perform.

What Develops?

In the preceding pages, I have tried to build the case that infants and adults

share a system for maintaining object representations in short-term memory.

This system is capacity limited and can only store representations of three to

68 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 68

four individuals at any one time. Once stored in short-term memory, infants

and adults can perform a range of computations over these representations.

For example, we have begun to understand the ways in which infants and

adults compute quantity, both discrete and continuous, over object arrays. In

addition, both infants and adults appear able to reorganize representations of

individuals into a hierarchical structure. This chunking of object representa-

tions provides a link between limited-capacity short-term memory and the

greater storage capacity of long-term memory, and has been shown to enable

both infants and adults to overcome the three-to-four-item limit on simulta-

neous representation. In all of these ways, the representations stored by and

the computations performed by infants and adults are strikingly similar.

Where, then, is the development in short-term memory? While I have

focused on the respects in which short-term memory may be continuous over

development, there may also be important ways in which early versus mature

short-term memory differs. I now point to some existing research, some of

which is addressed by other chapters in this volume, as well as to avenues

for future investigation of short-term memory development.

First, we have already explored some possible differences in short-term

memory computations in terms of bottom-up versus top-down processing.

Continuity of Format and Computation 69

Identical Objects

Contrasting Objects

Represent

Number?

Represent

Total Extent?

√

(Expts. 3-5,

Feigenson Carey, &

Spelke, 2002)

(Expt. 2,

Feigenson, 2005)

(Expt. 1,

Feigenson, 2005)

(Expt. 2,

Feigenson Carey, &

Spelke, 2002)

FIGURE 3.5. Double dissociation between array heterogeneity and the

computation infants perform over the objects in the array. Reprinted from

Feigenson, L. (2005). A double dissociation in infants’ representation of

from Elsevier.

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 69

Adults can use semantic knowledge to endogenously initiate the chunking of

representations held in short-term memory, and have volition over which

dimension of quantity to represent. It remains unclear whether infants also

have this ability. To date, infants’ computations over object arrays appear to

be driven from the bottom up by perceptual information present in the array

itself. Even if new research finds that infants can initiate the chunking of an

object array in a top-down fashion (for example, by using conceptual knowl-

edge of animals versus vehicles to parse an array into these two categories),

the comparative richness of adult knowledge about the world will likely be

reflected in developmental differences. If infants do have some top-down con-

trol over the chunking of object arrays, this control will almost certainly

increase over time. As they come to refine their knowledge of object kinds and

categories, infants may gain more ways to parse arrays into chunks and there-

fore gain more avenues for motivating top-down chunking. Exploring devel-

opmental changes in the top-down versus bottom-up execution of short-term

memory computations is a promising direction for future research.

In addition, much remains to be understood about the nature of infants’

and adults’ capacity limits. Although both groups appear able to store up to

three to four items in short-term memory, information capacity within each

of these three to four available slots is probably not fixed. For example, rep-

resenting three very complex objects with many features and articulated

parts may impose a higher informational load than representing three simple

geometric shapes. Alvarez and Cavanaugh (2004) measured this load empir-

ically using a change detection task, and confirmed that the number of

objects adults can store depends on the objects’ complexity. Adults were able

to maintain a larger number of items in visual short-term memory when

those items were simple colored squares than when the items were more

complex letters or shapes. These results show that although short-term

memory can store a maximum number of about four items, the information

load of the array can significantly reduce this capacity. This question has yet

to be systematically explored in infants, and it raises the possibility that in-

fants and adults may differ in the amount of information they can store in

each memory slot. Can infants represent multiple features of three to four

complex objects, or are they limited to representing just a few salient prop-

erties? Systematically manipulating object complexity will help characterize

the subtle limits of short-term memory development.

Finally, while the eventual upper limit on short-term memory appears

fixed at three to four items, infants’ memory capacity may not reach this limit

for some time. In their change detection experiments, Ross-Sheehy et al.

(2003) found that 10- and 13-month-old infants detected a changing item

contained within a three- or four-item array. But 4- and 6.5-month-old in-

fants could only detect a change with a one-item array, failing with arrays of

two and three. A similar pattern has been obtained by Kaldy and Leslie (in

press), using a quite different paradigm in which infants are asked to track

the shape of three-dimensional objects that move behind occluding screens.

70 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 70

Based on their results, Kaldy and Leslie suggest that 6.5-month-old infants’

short-term memory capacity is limited to just one slot. These findings raise

the possibility that the capacity of short-term memory increases over the first

year of life and reaches asymptote by 10 to 12 months.

This interesting developmental proposal may conflict with previous find-

ings that even by 5 months, infants can successfully track and remember at

least two hidden objects at a time (Koechlin et al., 1997; Simon et al., 1995;

Wynn, 1992). Ross-Sheehy et al. (2003) suggested that these previous results

might not be tapping short-term memory and that, because infants view

objects over much longer durations than in the change detection task (several

seconds, compared with 500 milliseconds), long-term memory might also be

involved. This issue merits deeper exploration. The three-to-four-item limit is

observed in tasks involving a wide range of durations, from 500 milliseconds

to 30 seconds or more. Does this commonality implicate a single system of

memory representation encompassing a wide span of durations? Might there

be multiple levels of memory storage that are all limited by a single bottleneck

on information processing?

These questions return us to the issues raised by Cowan’s (2001) contro-

versial proposal regarding the distinctions between attention, short-term

memory, and long-term memory. The developmental evidence reviewed

here does not provide definitive answers. Nonetheless, characterizing in-

fants’ memory—both its capacities and its limitations—may offer a window

into understanding these systems and the interactions between them. As we

identify the continuities and discontinuities in representational ability across

the life span, we add to the emerging portrait of memory development over

time.

Notes

1. Previous work (Feigenson, Carey, & Hauser, 2002; Scholl & Leslie, 1999;

Simon, 1997; Uller et al., 1999) has suggested that infants and adults share a

system that is dedicated to tracking objects per se, and to storing information

about their properties. This “object-file” system enables the creation of a

token, or file, that stands for an object in the world and allows it to be repre-

sented over changes in spatial location or changes in properties (Kahneman,

Treisman, & Gibbs, 1992). While such a system may indeed be in place

throughout development, here I make the more general claim that infants and

adults share a system for representing discrete items in short-term memory.

These items may be objects, but may also be nonobject entities such as sounds

or events that are perceived in any sensory modality.

2. Presenting the cue less than 1 second after the array disappeared

resulted in much higher capacity limits, which Sperling took as evidence for

the existence of a very short-lived, iconic memory store. Iconic memory,

which lasts for less than a second, is distinct from the short-term memory that

is the focus of the present discussion.

Continuity of Format and Computation 71

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 71

3. Note that this method does not allow us to be certain of the exact

number of objects infants represented in the box. For example, for infants who

saw three objects hidden, retrieved one of them, and then continued to search,

infants may have believed there to be exactly two objects still remaining.

Alternatively, it is possible that they represented just one more object in the

box or an unspecified number of objects remaining in the box.

References

Alvarez, G. A., & Cavanaugh, P. (2004). The capacity of visual short-term

memory is set both by visual information load and by number of objects.

Psychological Science, 15, 106–111.

Alvarez, G. A., & Franconeri, S. L. (2005). How many objects can you track?

Paper presented at the annual meeting of the Vision Sciences Society,

Sarasota, FL.

Baillargeon, R. (1986). Representing the existence and location of hidden

objects: Object permanence in 6- and 8-month old infants. Cognition, 23,

21–41.

Baillargeon, R., Spelke, E. S., & Wasserman, S. (1985). Object permanence in

five-month-old infants. Cognition, 20, 191–208.

Carey, S. (2004). Bootstrapping and the origins of concepts. Daedalus, 133,

59–68.

Chase, W. G., & Ericsson, K. A. (1981). Skilled memory. In J. R. Anderson

(Ed.), Cognitive skills and their acquisition (pp. 141–189). Mahwah, NJ:

Erlbaum.

Cheries, E. W., DeCoste, C., & Wynn, K. (2003). Number Not Area: Infants

Use Property Contrasts for Quantifying Objects. Poster presented at the

Society for Research in Child Development, Tampa, FL.

Clearfield, M. W., & Mix, K. S. (1999). Number versus contour length in

infants’ discrimination of small visual sets. Psychological Science, 10(5),

408–411.

Clearfield, M. W., & Mix, K. S. (2001). Infants’ use of area and contour length

to discriminate small sets. Journal of Cognition and Development, 2,

243–260.

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsid-

eration of mental storage capacity. Behavioral and Brain Sciences, 24,

87–185.

Diamond, A. (1990). Rate of maturation of the hippocampus and the devel-

opmental progression of children’s performance on the delayed non-

matching to sample and visual paired comparison tasks. Annals of the

New York Academy of Sciences, 608, 637–676.

Ericsson, K. A., Chase, W. G., & Faloon, S. (1980). Acquisition of a memory

skill. Science, 208, 1181–1182.

Feigenson, L. (2005). A double dissociation in infants’ representation of

object arrays. Cognition, 95, B37–B48.

Feigenson, L., & Carey, S. (2003). Tracking individuals via object-files:

Evidence from infants’ manual search. Developmental Science, 6, 568–584.

Feigenson, L., & Carey, S. (2005). On the limits of infants’ quantification of

small object arrays. Cognition, 97, 295–313.

72 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 72

Feigenson, L., Carey, S., & Hauser, M. (2002). The representations underly-

ing infants’ choice of more: Object files versus analog magnitudes.

Psychological Science, 13(2), 150–156.

Feigenson, L., Carey, S., & Spelke, E. (2002). Infants’ discrimination of num-

ber vs. continuous extent. Cognitive Psychology, 44(1), 33–66.

Feigenson, L., Dehaene, S., & Spelke, E. S. (2004). Core systems of number.

Trends in Cognitive Sciences 8(7), 307–314.

Feigenson, L., & Halberda, J. (2004). Infants chunk object arrays into sets of

individuals. Cognition, 91, 173–190.

Feigenson, L., & Halberda, J. (in preparation). Computations made over set

representations.

Fenson, L., Dale, P. S., Reznick, J. S., Bates, E., Thal, D., Pethick, S. J., et al.

(1994). Variability in early communicative development. Monographs of

the Society for Research in Child Development, 242.

Halberda, J., Simons, D., & Wetherhold, J. (2006). You can never attend to

more than three items at once: Gestalt grouping principles explain

changes in capacity. Manuscript submitted for publication.

Halberda, J., Sires, S. F., & Feigenson, L. (2006). Multiple spatially overlapped

sets can be enumerated in parallel. Psychological Science, 17(7), 572–576.

Hood, B., & Willats, P. (1986). Reaching in the dark to an object’s remem-

bered position: Evidence for object permanence in 5-month-old infants.

British Journal of Developmental Psychology, 4, 57–65.

Kahneman, D., Treisman, A., & Gibbs, B. (1992). The reviewing of object

files: Object-specific integration of information. Cognitive Psychology,

24(2), 175–219.

Kaldy, Z., & Leslie, A. M. (2003). Individuation of objects in 9-month-old

infants: Integrating “what” and “where” information. Developmental

Science, 6(3), 360–373.

Kaldy, Z., & Leslie, A. M. (2005). A memory span of one? Object identifica-

tion in 6.5-month old infants. Cognition, 97(2), 153–157.

Koechlin, E., Dehaene, S., & Mehler, J. (1997). Numerical transformations in

five month old human infants. Mathematical Cognition, 3, 89–104.

James, W. (1890). The principles of psychology. New York: Henry Holt.

Lipton, J. S., & Spelke, E. S. (2003). Origins of number sense: Large number

discrimination in human infants. Psychological Science, 15, 396–401.

Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for

features and conjunctions. Nature, 390, 279–281.

McCrink, K., & Wynn, K. (2004). Large-number addition and subtraction in

infants. Psychological Science, 15, 776–781.

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.

Noles, N., Scholl, B. J., & Mitroff, S. R. (2005). The persistence of object file

representations. Perception and Psychophysics, 67(2), 324–334.

Pylyshyn, Z. W. (1994). Some primitive mechanisms of spatial attention.

Cognition, 50, 363–384.

Pylyshyn, Z. W. (2001). Visual indexes, preconceptual objects, and situated

vision. Cognition, 80, 127–158.

Pylyshyn, Z. W., & Storm, R. W. (1988). Tracking multiple independent targets:

Evidence for a parallel tracking mechanism. Spatial Vision, 3(3), 179–197.

Continuity of Format and Computation 73

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 73

Ross-Sheehy, S., Oakes, L. M., & Luck, S. J. (2003). The development of visual

short-term memory capacity in infants. Child Development, 74, 1807–1822.

Scholl, B. J., & Leslie, A. M. (1999). Explaining the infant’s object concept:

Beyond the perception/cognition dichotomy. In E. Lepore & Z. Pylyshyn

(Eds.), What is cognitive science? (pp. 26–73). Oxford, UK: Blackwell.

Shinskey, J. L., & Munakata, Y. (2003). Are infants in the dark about hidden

objects? Developmental Science, 6, 273–282.

Simon, H. A., & Chase, W. G. (1973). Skill in chess. American Scientist, 61,

393–403.

Simon, T., Hespos, S. J., & Rochat, P. (1995). Do infants understand simple

arithmetic? A replication of Wynn (1992). Cognitive Development, 10,

253–269.

Simon, T. J. (1997). Reconceptualizing the origins of number knowledge: A

“non-numerical” account. Cognitive Development, 12, 349–372.

Sperling, G. (1960). The information available in brief visual presentations.

Psychological Monographs, 74.

Trick, L., & Pylyshyn, Z. (1994). Why are small and large numbers enumer-

ated differently? A limited-capacity preattentive stage in vision. Psycho-

logical Review, 101(1), 80–102.

Uller, C., Huntley-Fenner, G., Carey, S., & Klatt, L. (1999). What representa-

tions might underlie infant numerical knowledge? Cognitive Develop-

ment, 14, 1–36.

Wynn, K. (1992). Addition and subtraction by human infants. Nature, 358,

749–750.

Xu, F. (2003). Numerosity discrimination in infants: Evidence for two sys-

tems of representation. Cognition, 89, B15–B25.

Xu, F., & Spelke, E. S. (2000). Large number discrimination in 6-month old

infants. Cognition, 74, B1–B11.

Xu, F., Spelke, E. S., & Goddard, S. (2005). Number sense in human infants.

Developmental Science, 8, 88–101.

74 Short-Term or Working Memory

Oakes_Ch_03.qxd 11/8/2006 3:47 PM Page 74