Language, space, and the development of
cognitive ¯exibility in humans: the case of two
spatial memory tasks
Linda Hermer-Vazquez*, Anne Moffet, Paul Munkholm
Department of Psychology and Cognitive Studies Program, Cornell University, Ithaca, NY 14850, USA
Received 10 February 2000; received in revised form 5 June 2000; accepted 23 September 2000
Prior experiments have shown that young children, like adult rats, rely mainly on informa-
tion about the macroscopic shape of the environment to reorient themselves, whereas human
adults rely more ¯exibly on combinations of spatial and non-spatial landmark information.
Adult rats have also been shown to exhibit a striking limitation in another spatial memory
task, movable object search, again a limitation not shown by human adults. The present
experiments explored the developmental change in humans leading to more ¯exible,
human adult-like performance on these two tasks. Experiment 1 identi®ed the age range of
5±7 years as the time the developmental change for reorientation occurs. Experiment 2
employed a multiple regression approach to determine that among several candidate
measures, only a speci®c language production measure, the production of phrases specifying
exactly the information needed to solve the task like adults, correlated with the reorientation
performance of children in this age range. Experiment 3 revealed that similar language
production abilities were associated with more ¯exible moving object search task perfor-
mance. These results, in combination with ®ndings with human adults, suggest that language
production skills play a causal role in allowing older humans to construct novel representa-
tions rapidly, which can then be used to transcend the limits of phylogenetically older cogni-
tive processes. q2001 Elsevier Science B.V. All rights reserved.
Keywords: Language; Space; Cognitive ¯exibility; Humans; Spatial memory tasks
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 263
Cognition 79 (2001) 263±299
0010-0277/01/$ - see front matter q2001 Elsevier Science B.V. All rights reserved.
The experiments described in this article have been reported previously at conferences in poster form
(Hermer, 1994, 1998) and in a published dissertation (Hermer, 1997a).
* Corresponding author. Present address: SUNY Health Science Center, Department of Physiology and
Pharmacology, Box 31, 450 Clarkson Avenue, Brooklyn, NY 11203, USA. Fax: 11-718-270-3103.
E-mail address: firstname.lastname@example.org (L. Hermer-Vazquez).
Over the course of the 100 000±150 000 year history of our species, Homo
sapiens sapiens, the knowledge possessed by individuals and cultures has exploded.
Faint beginnings of art and religion appear in tombs and artifacts from Africa 90 000
years ago (Holden, 1998). Within a few tens of thousands of years, cave art, sculp-
ture, stone and metal work, and many varieties of tool use ¯ourished (Appenzeller,
1998). Shortly before the onset of the Agricultural Revolution about 10 000 years
ago, group size and cultural organization expanded greatly (Balter, 1998), and in the
last 200 years alone we have undergone the Industrial and Information Revolutions.
In so doing, humans have de®ned and solved many novel problems for which
biology had not explicitly prepared us, such as inventing the wheel, the internal
combustion engine and the computer, and creating the Internet. What cognitive
capacities gave us the ¯exibility to invent these technologies and master the systems
of knowledge that underpin them?
Assuming that cognition follows general principles of biological evolution, biol-
ogy offers some guidance as to how to think about these special cognitive mechan-
isms. In general, there is tremendous conservation of biological processes, and
phylogenetically newer mechanisms often are layered on top of older ones (Ridley,
1993). For example, most cells evolved the ability to perform cellular respiration
once oxygen became abundant in the environment, but virtually all living cells still
also perform glycolysis, the ancient energy extraction solution (Alberts et al., 1992).
Moreover, recent innovations often develop later in ontogeny than do older mechan-
isms, so as not to disrupt older mechanisms and the bene®cial ways in which they
mesh with other emerging traits (Ridley, 1993). For instance, cartilage appeared in
®shes long before bones, and cartilage develops in bony ®shes early in life, later
replaced by bone (Ridley, 1993).
Evolution that proceeds in this way is called terminal addition (Ridley, 1993).
Although evolution often progresses in other ways, terminal addition occurs often
enough to suggest one logic for studying the emergence of human-speci®c cognitive
traits: ®nd a trait for which young children show the phylogenetically older and more
common mechanism but for which human adults show distinctive ¯exibility, and
then study the developmental change in depth. To pursue this goal, it is important to
choose a research area about which much is known so that the extension of the core
trait by the human-speci®c trait can be understood in detail.
With these considerations, we focused on the research area of navigation and
spatial memory. Since the discovery of putative `cognitive maps' 50 years ago
(Tolman, 1948) and the discovery of place cells of the hippocampus 30 years ago
(O'Keefe & Dostrovsky, 1971), there has been an explosion of behavioral and
neurobiological research in this area. Moreover, Biegler and Morris (1993) and
Cheng (1986) reported striking limitations shown in the spatial memory abilities
of adult rats, which do not appear to be shown by adult humans. The limitation
Cheng discovered concerned a process called spatial reorientation. When mammals
move about, they normally update their position and heading by path integration,
using vestibular, motor feedback, and optical ¯ow signals to compute the extent of
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299264
their own motion and to update their position (Etienne, Maurer, & Seguinot, 1996;
Gallistel, 1990; McNaughton et al., 1996; Mittelstaedt & Mittelstaedt, 1980). Path
integration by itself, however, accumulates error over time and thus requires peri-
odic recalibration with information from spatial memory (McNaughton et al., 1996).
When path integration is completely disabled through a disorientation process, the
contents of spatial memory can be studied without the confound of oriented spatial
Accordingly, Cheng (1986) placed adult rats in a foraging task in which a hidden
food location was partly speci®ed by the three-dimensional geometry of the testing
cage and fully speci®ed by distinctive odors, patterns and brightness placed at
different locations in the cage. The rat was given three morsels of food at a particular
location, determined by the experimenter, and then was removed from the cage. The
cage was rotated by a random amount so that it was misaligned relative to the rat's
internal sense of orientation, and the rat was returned to the cage and allowed to dig
for more food, presumably by using their memory of the location at which they had
just received food. Cheng assumed that the disoriented rats would use the currently
perceived environment and their spatial memory ®rst to reorient themselves, and
then to retrieve the food's position from their cognitive map, an assumption which
received explicit support from behavioral studies of young children (Hermer, 1997b)
and indirect support from neurophysiological studies of rats (Knierim, Kudrimoti, &
McNaughton, 1995). Rats had repeatedly been shown to use odors, brightness and
patterns as discriminative cues (e.g. Suzuki, Augerinos, & Black, 1980). However,
in contrast to Cheng's expectations, the rats relied heavily on the three-dimensional
geometry of the cage, but not on the other, `non-geometric' cues to reorient them-
selves (Fig. 1). Although the search patterns of Cheng's rats suggested some residual
reliance on non-geometric cues, that suggestion was rejected in further research by
Margules and Gallistel (1988), who tested rats in a similar task but using a more
effective disorientation procedure. Fully disoriented rats were found to rely exclu-
sively on the geometry of the cage to guide their search, at least under circumstances
similar to Cheng's.
Other research has made clear that rats' spatial memory is open to non-geometric
information as well, and under some circumstances rats rely on non-geometric cues
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 265
Fig. 1. Testing environments and results for an experiment by Cheng (1986). Search results are collapsed
across search locations and across the three subjects in each experiment, with standard errors in parenth-
eses. Non-geometric patterns on corner panels represent both the unique visual pattern and the unique
scent on each panel. Redrawn from Cheng (1986).
to reorient themselves (e.g. in the aversive Morris water maze; Dudchenko, Good-
ridge, Seiterle, & Taube, 1997). In many circumstances, however, such as in appe-
titive tasks in which the environment contains distinctive geometry and path
integration is completely disabled, rats rely primarily on geometric information to
reorient themselves. These limitations provided the circumstances for probing
humans' possibly more ¯exible performance and the developmental changes that
would give rise to it.
Hermer and Spelke (1994, 1996) tested young children's and adults' reorientation
capacities in a situation similar to Cheng's. First they familiarized human adults
with a rectangular room with a red panel in each corner behind which an object
could be hidden. Subjects saw an object being hidden in one of the corners, they
underwent a disorientation procedure, and they were asked to locate the object. In
one condition, the walls of the room were entirely white and the room was devoid of
other landmarks, so that the room's geometry provided the only information to guide
adults' search. As in Cheng's studies, the rotational symmetry of the rectangular
room about its vertical axis failed to distinguish the correct search location from its
1808rotational equivalent, and so disoriented adults who were able to reorient by the
room geometry would be expected to locate the target only 50% of the time. Indeed,
this is what happened. In the all-white condition, adult subjects split their search
evenly between the two geometrically appropriate locations. This pattern of results
indicated that they were fully disoriented and that they used the room's geometry as
a cue to orientation and object location.
In a second condition (conducted ®rst for half of the subjects), one of the cham-
ber's short walls was covered with a bright blue cloth. This landmark broke the
symmetry of the room and would permit subjects to locate the object on 100% of
trials, if they could encode the object's position and their own orientation using it as
an indirect cue. This condition was run with a similar procedure, starting with
subjects being familiarized with the features of the room before the trials
commenced. In this condition, and in contrast to Cheng's and Margules and Gallis-
tel's rats, human adults con®ned their search to the correct location, indicating that
they used the non-geometric cue of the wall's color as an indirect cue to orientation
and/or to the object's location (Fig. 2). In this condition human adults therefore
showed more ¯exibility than adult rats in their reliance on conjunctions of spatial
and non-spatial information to solve the task.
Hermer and Spelke (1994, 1996) further tested children of 18±24 months of age in
the same tasks. In the all-white condition, young children performed like adult rats
and human adults in that they relied on the geometry of the environment to reorient
themselves. Like adult rats and unlike human adults, however, young children failed
to use the blue wall to reorient themselves (Fig. 3). In the condition with one blue
wall, children continued to split their search evenly between the two geometrically
appropriate locations, even though some children spontaneously commented on the
presence of the blue wall, indicating that they had noticed it. Separate analyses of
their data showed no ability to rely on the blue wall for reorientation (Hermer,
Further experiments undertaken with children ranging in age between 18 months
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299266
(Hermer & Spelke, 1994, 1996) and 4 years (Hermer, 1997b) revealed that children
encode the non-geometric properties of the test environments and can use these
properties to solve other memory tasks, but con®rmed that such children do not
reorient using this information, at least with the present task. The process of spatial
reorientation thus exhibited all the properties cited above as a candidate for study of
the factors underlying human adult cognitive ¯exibility. Adult rats showed a striking
limitation in the kind of information they could use to solve reorientation tasks.
Human adults, in contrast, showed more ¯exibility than adult rats in the information
they could use. And human adults' distinctive ¯exibility did not emerge in children
until relatively late in postnatal development (after the age of 4 years), making
possible a detailed study of the developmental mechanisms responsible for the
emergence of the abilities found in human adults.
In the present article, we report three studies of the development of ¯exible
reorientation in older children. Experiment 1 was undertaken to identify the approx-
imate age at which the transition to adult-like abilities for reorientation occurs.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 267
Research by Cheng (1986), Gallistel (1990) and Vallortigara et al. (1991) shows that with training,
disoriented rats and domestic chicks can form direct associations between non-geometric landmarks and
use these associations to appear to solve some reorientation tasks. Over the period of extensive training,
however, neither rats nor domestic chicks developed the ability to use the non-geometric landmarks as
indirect spatial cues, signifying that they were likely not reorienting when they used them as direct cues.
However, as previously stated, other research indicates that non-human species can, under some circum-
stances, reorient by non-geometric information. In the aversive Morris water maze, rats appear to reorient
themselves with non-geometric cues (Dudchenko et al., 1997). Moreover, Gouteux, Thinus-Blanc, and
Vauclair (submitted for publication) provide evidence that disoriented rats and monkeys can use non-
geometric information to locate an object if they receive extensive training with the non-geometric
landmark and have the opportunity to learn, over an extended time period, about the relation of the
landmark to the object. Further research by Miller et al. (in preparation) provides evidence that cotton-
top tamarins who are disoriented come rapidly to use a non-geometric landmark to locate a hidden object,
when tested under conditions similar to those used with human children and adults. Because the monkeys
had received months of training in a similar landmark task, however (Delpolyi and Hauser, unpublished
data), no result has yet conclusively shown that non-human primates show the spontaneous, ¯exible
indirect landmark use found with human adults' reorientation.
Fig. 2. Search (and standard errors) of human adults in a rectangular room with one blue wall. C,
geometrically and non-geometrically correct corner; N, near, non-geometrically correct but geometrically
wrong corner; R, rotationally equivalent corner which was geometrically correct but non-geometrically
wrong; F, far corner that was geometrically and non-geometrically wrong. Redrawn from Hermer and
Experiment 2 was undertaken to investigate correlations between the onset of ¯ex-
ible performance of this task and other developing cognitive capacities. Experiment
3 investigated whether the same cognitive capacities that correlated with ¯exible
reorientation also correlated with ¯exible performance of another spatial memory
task: the search for a movable, hidden toy that bore a constant relation to a movable,
visible landmark. In this task, as with reorientation, adult rats show a striking
limitation, immediately searching proximally to the moving landmark but requiring
hundreds of trials to learn to solve the task using precise spatial information in
combination with non-spatial information (Biegler & Morris, 1996). In contrast,
human adults combine the information together to solve the task correctly on
their ®rst trial (Hermer-Vazquez, Spelke, & Katsnelson, 1999).
2. Experiment 1
The ®rst step in determining which cognitive mechanisms underlay human adult
¯exibility for reorientation was to identify the approximate age at which children
began performing like adults in the reorientation task described above. A prior study
had shown that children aged 3±4 years continued to perform like younger children
and adult rats (Hermer, 1997b), and pilot work had indicated that the transition to
adult-like performance occurred sometime between the ages of 5 and 7 years. The
®rst purpose of this experiment was thus to con®rm that children of 3±4 years of age
would fail to conjoin geometric and non-geometric information to solve this task,
but that older children in the 6 year age range would combine these types of
information together to solve the task in the manner of human adults.
The second purpose of this experiment was to provide one test of whether subjects
who rely on non-geometric information in this task are truly reorienting by that
information. If children were reorienting using the blue wall, the age at which they
begin using it as a landmark in a direct spatial memory task where the object was
hidden directly behind it might also be the age at which they began using it as an
indirect cue. That is, if such subjects had placed the blue wall on a cognitive map
that could be used for reorientation, they should solve the direct and indirect tasks at
the same age. If in contrast, they either use it initially as a beacon (Gallistel, 1990)
only, or as part of a local spatial or propositional construct (e.g. `the toy is to the left
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299268
Fig. 3. Search (and standard errors) of children of 18±24 months of age in the same reorientation task as in
Fig. 2. Redrawn from Hermer and Spelke (1996).
of the blue wall', with the blue wall not located on a larger cognitive map), they
might solve the direct task before the indirect task. To test these possibilities, two
groups of children in the two age ranges above were tested in two `reorientation'
tasks, one with the blue wall as a direct cue to the hidden object's location and the
other with the blue wall as an indirect cue to object location.
Participants were eight males and eight females between the ages of 3.0 and 4.0
years (mean age 3.4 years) and ten males and six females between the ages of 5.4
and 6.0 years (mean age 5.9 years). Three additional children had to be omitted from
the sample and replaced due to experimenter or parent error (see below).
Subjects were tested in a 1.91 £1.22 £1.91 m rectangular chamber housed within
a larger experiment room with no windows or obvious sources of noise. The cham-
ber was composed of white felt fabric stretched onto a concealed wooden frame and
a padded ¯oor. A curtained opening to the left of one of the long walls (as one faces
it from the outside) permitted entry into the room without breaking its symmetry;
when not in use, this opening was sealed with Velcro. Four indistinguishable
23 £122 cm red panels composed of felt on a concealed wooden frame with a
loose fabric curtain at the bottom stood in the room's four corners. In both condi-
tions, a bright blue 1.22 £1.91 m piece of fabric was attached with Velcro to the
shorter wall opposite the side of the chamber with the door such that it covered that
wall completely. The fabric on the white walls was not permanently sealed to the
¯oor and the bottom of the fabric could be lifted so that a toy could be hidden behind
it. The room was illuminated from above by four 25 W lights, one in the top center of
each wall. A video camera suspended from the center of the room's ceiling provided
an overhead view of the experiment. A central overhead white noise generator
prevented subjects from maintaining their orientation through the use of any
sound beacon. A ring of keys or a small toy chosen by the child in the waiting
room served as the object for which the subject searched.
Participants at both ages received three to four codable trials in the Direct Land-
mark condition and three to four codable trials in the Indirect Landmark condition.
In the Direct Landmark condition, the hiding location was constant within and
across children (directly behind the blue fabric). In the Indirect Landmark condition,
the hiding location was constant within each child but was counterbalanced across
corners between children. The order of the sessions (Direct Landmark condition ®rst
or second) was also counterbalanced between subjects.
In the Direct Landmark condition, each child entered the room with the experi-
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 269
menter, who then sealed the door from the inside. A parent sometimes accompanied
a child into the testing room, depending on the preference of the child. The experi-
menter pointed out the white and blue short walls to the child and showed him or her
how a toy could be hidden behind either of them. After this demonstration, she
proceeded to hide a toy directly behind the center bottom of the blue fabric while the
child watched. The experimenter asked the child to point to where the object had
been hidden. Once this had occurred successfully, either the experimenter lifted the
child, covered his eyes, and turned him ®ve full times, or the child spun himself with
his eyes closed, ®ve full times. When a child spun himself, the experimenter walked
around him slowly so that her position could not be used as a cue to the object's
location. When a parent also was in the room, the parent moved around as well. The
experimenter set the child down or asked him or her to stop spinning, facing a
predetermined random direction. He was then told to open his eyes and retrieve
The overhead video recorder and the experimenter from inside the chamber
recorded the hiding location, the subject's facing position, and his ®rst (and subse-
quent, if his ®rst search was incorrect) search locations. Once this had occurred, the
next trial began as before, until four trials had been completed. A small number of
subjects only received three trials due to fussiness.
In the Indirect Landmark condition, each child entered the room with the experi-
menter and the experimenter sealed the door as in the Direct Landmark condition.
She pointed out the two short walls but then turned to the corner pillar that would be
used as that child's hiding place and demonstrated how an object could be hidden
under the fabric at the bottom of the pillar. The child was asked to point to where the
toy had been hidden, as in the Direct Landmark condition, after which the disor-
ientation procedure, object search, and trial recording took place as above. Again,
nearly all children received four trials but a small number received only three trials
due to fussiness.
2.1.5. Coding and analyses
All searches for the object were coded from the video record by assistants who
were aware of the purpose of the experiment but who were blind to the hidden
object's location. Coders considered a child to have searched for the object when-
ever he was judged to have pointed to a panel after disorientation, regardless of
whether the object was retrieved at that corner. To make this judgment, one experi-
menter cued the videotape to the point immediately after the hiding of the object,
and the other experimenter judged the direction in which the child faced at the end of
disorientation and the location of each of the child's searches for the object. This
procedure ensured that the coder of the child's search performance was blind to the
hidden object's location.
The principal analyses focused on the location of the child's ®rst search on each
search trial. As in our previous experiments (Hermer, 1997b; Hermer & Spelke,
1994, 1996; Hermer-Vazquez et al., 1999), the search location was coded along two
dimensions. For the Indirect Landmark condition, it was coded as `geometrically
appropriate' if the child searched either at the correct corner (C) or at its rotational
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299270
equivalent (R) and as `geometrically inappropriate' otherwise (corners N and F), and
it was coded as `proximal' if the child searched either at C or at the corner nearest to
C (N) and as `distant' otherwise (R and F). (Corners C and N both corresponded in
color to the correct corner ± either both were entirely white or both were half white
and half blue ± and so the proximal/distant distinction corresponded to a distinction
between search at appropriately versus inappropriately colored corners.) Finally, we
coded whether the search took place at the absolutely correct corner, which requires
a conjunction of correct geometric and correct proximal responding.
Coding of the Direct Landmark condition took place similarly. A search at the
correct wall was coded as absolutely correct. A search at either the correct wall (C)
or the opposite, short wall (R) was coded as geometrically appropriate, whereas a
search at any other location was coded as geometrically inappropriate (GI). Finally,
a search at either the correct wall or either of the two corners adjoining the correct
wall (this search pattern took place on a few trials) was coded as proximal, whereas a
search at any other wall or corner was coded as distant.
Analyses were conducted in three stages and were performed separately on chil-
dren's tendency to use geometric, proximal, and absolutely correct information. For
each of the three types of information, the ®rst stage consisted of binomial tests on
the ®rst-trial data from each condition to assess how children performed prior to any
experience with the tasks. The binomial tests analyzed whether each age group
performed above chance in their use of each of these three types of information.
Chance was de®ned as 50% for geometric and proximal responding, because in each
of those cases two of the four possible room locations were correct. For absolutely
correct responding, chance was also de®ned as 50%, because previous research had
shown that younger children's responses were concentrated in the two geometrically
correct locations and were evenly split between them (Hermer & Spelke, 1994,
1996). Next, to analyze data across trials, which were not independent, t-tests
were performed. Paired t-tests compared the amount of search using each type of
information, within each age and condition. Unpaired t-tests compared the use of
each type of information in each condition between the two age groups. Finally,
ANOVAs were performed on patterns of search at geometrically correct, proximally
correct, and absolutely correct locations to test for effects of speci®c factors on
children's search. The ANOVAs included the within-subjects factors of condition
(two: Direct Landmark and Indirect Landmark) and trial (four) and the between-
subjects factors of sex (two) and age (two: 3.0±4.0 years and 5.4±6.5 years). Type III
statistics and results were used, with each independent variable's contribution eval-
uated with other factors accounted for.
Fig. 4 presents children's overall search patterns in this experiment, broken down
by age and condition. It is apparent from visual inspection that children at both ages
searched the absolutely correct location with high frequency in the Direct Landmark
condition (Fig. 4a,b). In contrast, younger subjects split their search roughly evenly
between the two geometrically correct corners in the Indirect Landmark condition,
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 271
whereas older subjects tended to search the absolutely correct corner (Fig. 4c,d). In
both conditions, moreover, children searched geometrically appropriate locations
more than geometrically inappropriate locations. We consider performance in the
two landmark conditions separately.
2.2.1. Direct Landmark condition
Data from the children's search on the ®rst trial mostly con®rmed their high
search accuracy at both ages in this condition. Both the younger and the older
children searched geometrically appropriate locations more often than geometrically
inappropriate ones (respective binomial values: P0:0003 and 0.0026). Younger
and older children also made use of proximal, landmark-based information in the
Direct Landmark condition (respective values: P0:0021 and 0.0003). Finally,
children in both age groups searched the absolutely correct location more often
than the geometrically correct, opposite location on the ®rst trial in the Direct
Landmark condition (both P0:0005).
The analysis of performance across all trials (Fig. 4a,b) indicated that subjects'
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299272
Fig. 4. Search in the Direct and Indirect Landmark conditions of Experiment 1 by children in each age
group. (a) 3±4-year-olds in the Direct Landmark task. (b) 5.5±6.5-year-olds in the Direct Landmark task.
(c) 3±4-year-olds in the Indirect Landmark task. (d) 5.5±6.5-year-olds in the Indirect Landmark task. For
the Direct Landmark task: C, geometrically and non-geometrically correct wall location; R, rotationally
equivalent, geometrically correct but non-geometrically wrong wall location; GI, geometrically inap-
propriate wall locations. A small percentage of search occurred in the corners for this condition, and
was coded as geometrically inappropriate and as proximal or distal depending on the corner's location
relative to the correct wall. For the Indirect Landmark task: C, geometrically and non-geometrically
correct corner; N, near, non-geometrically correct but geometrically wrong corner; R, rotationally equiva-
lent corner which was geometrically correct but non-geometrically wrong; F, far corner that was geome-
trically and non-geometrically wrong.
high performance in this condition persisted over trials. Younger and older children
searched geometrically appropriate locations more than geometrically inappropriate
ones in the Direct Landmark condition (respective paired values: t156:161, P
0:0001 and 25.98, P,0:0001). Unpaired t-tests comparing the search between the
two age groups suggested that older and younger children made equal use of
geometric information (jt30j ,1:2). Both younger and older children made signif-
icant use of proximal landmark information in the Direct Landmark condition
(respective values: t155:861 and 9.925, P,0:0001). Moreover, there was
no age difference in the use of proximal information in the Direct Landmark condi-
tion (unpaired jt30j ,1). Finally, children tended to search the absolutely correct
location in the Direct Landmark condition at both the younger (t157:064,
P,0:0001), and the older age (t159:797, P,0:0001), with no difference
between the groups in absolute search accuracy (unpaired jt30j ,1:1).
2.2.2. Indirect Landmark condition
On the ®rst search trial, both the younger and the older children searched geome-
trically appropriate locations more often than geometrically inappropriate ones
(both binomial P0:0106). At neither age did children reliably use proximal
information to guide their search on the ®rst trial (both binomial P.0:40). At
neither age did children search the correct location more than the geometrically
equivalent opposite corner (for younger children, binomial P0:50; for older
Across all trials (Fig. 4c,d), children at both ages continued to search geometri-
cally appropriate more than inappropriate locations (for younger children,
t153:890, P0:0014; for older children, t159:141, P,0:0001), but
older children used geometric information more consistently than younger ones
(t3022:155, P0:0394). The two age groups diverged more strongly in
their use of proximal information in the indirect task: younger subjects failed to
rely on this information (jt15j ,1), whereas older subjects made signi®cant use of
it (t152:735, P0:0313), producing a non-signi®cant trend toward greater use
of this information at the older age (unpaired t301:546, P0:1325). Younger
children failed to search the absolutely correct location more than the geometrically
identical, opposite location (jt15j ,1), whereas older children searched the
correct location more successfully (t152:931, P0:0103). A comparison
using an unpaired t-test con®rmed that the older children searched the absolutely
correct corner substantially more often than the younger ones (unpaired
2.2.3. Comparisons across conditions
To compare performance across the Direct Landmark and the Indirect Landmark
conditions and to assess how effects of trial and subject sex interacted with the
effects described above, ANOVAs were also performed for each of the three search
measures (geometry, proximity, and absolutely correct search). Because a total of
eight subjects in the two age ranges lacked a fourth testing trial in one or both
conditions, their data could not be included in these ANOVAs. For subjects' use
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 273
of geometric information, the ANOVA revealed signi®cant main effects of age
(F1;217:30, P0:0133), re¯ecting older subjects' superior use of geometric
information, sex (F1;216:07, P0:0225), with girls making more use of
geometric information, and condition (F1;214:87, P0:0385), with better
performance in the Direct Landmark condition. The analysis of subjects' use of
proximal, landmark-based information revealed main effects of age
(F1;215:23, P0:0327), with older children making more use of the land-
mark information, and condition (F1;2116:91, P0:0005), with more consis-
tent use of landmark information in the Direct Landmark condition. The analysis of
search at the absolutely correct location revealed the same main effects of age
(F1;217:04, P0:0148) and condition (F1;2121:28, P0:0002),
again re¯ecting more accurate search at the older age and in the Direct Landmark
condition. There were no trial-related effects.
At both age levels, children were strikingly successful at using non-geometric
information to locate a hidden object in the Direct Landmark task: when a toy was
hidden directly behind a blue wall, children searched for the toy behind that wall in
preference to all other locations. This ®nding indicates that children from 3 to 6.5
years of age consistently perceived and remembered the location of the blue wall
and used its location to guide their search for the object.
Children's success in the Direct Landmark task does not imply, however, that the
children were able to reorient themselves with non-geometric information, using the
blue wall to reestablish their own position and the position of all known locations
around them after losing track of their own orientation. If children had used the blue
wall to reorient themselves, then they should have searched for the hidden toy in the
correct relation to the blue wall in the Indirect Landmark condition as well as in the
Direct Landmark condition. In fact, however, the search in relation to the landmark
was less consistent in the Indirect Landmark condition at both ages, and it did not
differ from chance at the younger age. These ®ndings suggest that children located
the object in the Direct Landmark task not by reorienting themselves in relation to
the blue wall but by forming a direct association between the blue wall and the
hidden object. Children, like rats and other mammals, have systems for forming
direct associations between a stimulus, a pleasurable outcome, and the motor
response they must make to obtain the reward (human infants, e.g. Rovee-Collier,
Borza, Adler, & Boller, 1993; adult rats, e.g. McDonald & White, 1993).
The last conclusions from Experiment 1 concern the developmental change in
performance on the Indirect Landmark task. The younger group of subjects failed to
rely on the blue wall in this task and showed the same pattern of geometry-based
search as 18±24-month-old subjects repeatedly had in our prior studies. This
tendency occurred despite the fact that the non-geometric cue in this task was
consistently present, in contrast to our earlier experiments in which the blue wall
covering was present in one condition and absent in the other. However, it should be
noted that while some experimenters have replicated these ®ndings (J. Stedron, pers.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299274
commun.; Wang, Hermer-Vazquez, & Spelke, 1999), others have not (Learmonth,
Newcombe, & Huttenlocher, 1998).
In contrast to the younger children, children who were 5.4±6.5 years of age
successfully located the hidden object by conjoining non-geometric and spatial
information and searching in the corner with the correct relationship to the blue
wall. Although such children did not succeed on this task on their ®rst trial, and
although they never performed as accurately as on the Direct Landmark task, they
succeeded over the course of all trials, appearing to learn extremely rapidly to solve
the task more ¯exibly. They may have either conjoined geometric information and
information about the blue wall, or conjoined another form of spatial information
such as propositionally-based sense information (`to the right of') with information
about the blue wall. In either case 5.4±6.5-year-old children appeared to learn very
rapidly to conjoin information across two domains of knowledge, namely spatial
knowledge and knowledge of object properties (Goodale, 1995; Simons, 1996;
Ungerleider, Courtney, & Haxby, 1997), for use in object search if not also for
The developmental change observed in the Indirect Landmark condition raises a
question: how do young children become more ¯exible in their reorientation or
object search? What allows them to conjoin spatial and non-geometric information
together so quickly, when adult rats failed to learn to do so across hundred of trials
(Cheng, 1986; Gallistel, 1990)? The next study began to address these questions by
investigating correlates of older children's success on the Indirect Landmark task.
3. Experiment 2
Experiment 2 employed a multiple regression approach to test for possible devel-
opmental correlates of the change from less ¯exible, geometry-based performance to
more ¯exible, geometry-plus-landmark-based performance on Cheng's reorienta-
tion task. The dependent variable was performance on a task similar to that in the
Indirect Landmark condition of Experiment 1, except that it took place in a square
room with one red wall and three white walls. The independent variables whose
associations with reorientation performance were tested were age, performance on a
standardized test of nonverbal intelligence (TONI-II), digit span, spatial memory
span, reorientation performance in an entirely white square room (a task assessing
the effectiveness of the disorientation procedure), comprehension and production of
phrases involving `to the left/right of', and comprehension and production of phrases
involving other spatial prepositions such as `above' and `behind'.
If conjunctive reorientation or object search (henceforth `reorientation', although
the ®nding from Experiment 1 that success at the Direct Landmark task did not
automatically lead to success at the Indirect Landmark task indicates that older
children might not be reorienting themselves in the Indirect task) depends on
improvements in general cognitive abilities or on an increase in knowledge about
the world, then landmark-based reorientation should correlate with age and intelli-
gence. If successful reorientation depends on improvements in general processing
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 275
capacity (e.g. Kuhn, 1992), then landmark-based reorientation should correlate with
digit span, spatial memory span, and perhaps age and intelligence. If older children
improve their performance by learning to resist becoming disoriented, then perfor-
mance in the reorientation task with the red wall should correlate with performance
in the all-white room. If ¯exible spatial performance is related in a general way to
the development of spatial language or of the spatial concepts that are expressed in
spatial language, then performance on the landmark-based reorientation task should
correlate with all of the measures of language comprehension or production. Finally,
if successful landmark-based reorientation depends on a process of using language
actively to describe the spatial relationship of the hidden object to the visible land-
mark, then landmark-based reorientation should correlate speci®cally with the
production of spatial expressions involving `right' and `left'.
In order for the regression experiment to be meaningful, it was necessary to test
children at an age at which some were likely to succeed at, and some to fail, the
landmark-based reorientation task that provided the dependent measure. Among the
16 5.5±6.5-year-olds in Experiment 1, seven children performed correctly on two or
fewer of the trials in the Indirect Landmark condition, suggesting that there is
considerable variation in performance at this age. Accordingly, Experiment 2 was
conducted with children in roughly the same age range.
A second set of requirements for the regression analysis was that the independent
measures have high validity and show suf®cient variability across children in the
sample. To maximize validity and variability, we chose standardized measures and
well-studied tasks to assess intelligence, verbal working memory, and spatial work-
ing memory, and we aimed for a level of dif®culty that would maximally differ-
entiate among the children.
Ten boys and 14 girls between the ages of 5.4 and 6.2 years (mean age 5.8 years)
took part in the experiment. Children were recruited as before and were paid $10 for
each of the two visits to the lab.
Children visited the lab on two occasions with their visits spaced less than 1.5
weeks apart. On the ®rst visit, children were tested on a digit span task, a moving
object search task whose results will not be reported here, a conjunction memory test
whose results also will not be reported here, a reorientation task in the square room
with one red wall, and a reorientation task in the square room with four white walls,
in that order. On the second visit, children were given an IQ test, a test of spatial
language production, a test of spatial language comprehension, and a visuospatial
memory span task, in that order. A detailed description of each task to be reported
22.214.171.124. Reorientation tasks. These tasks took place in a 1.87 £1.87 £1.91 m square
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299276
room. The room had three white walls and one red wall in the landmark-based
reorientation task and four white walls in the disorientation control task. Each
corner contained a red panel, described earlier, behind which a toy could be
hidden. A central overhead video and audio camera recorded all trials and a
central overhead white noise generator prevented subjects from remaining
oriented through the use of any sound beacon. Each child selected a small toy
from a box of toys that would ®t behind the corner panels as the object for which
he would search. At the beginning of the study, the child entered the experiment
room with the experimenter and the experimenter sealed the door from the inside,
making it invisible. The experimenter pointed out the red wall to the child and then
hid the toy behind one corner panel. The experimenter turned the child slightly and
asked him where the toy had been hidden; the rest of the trial only ensued if the child
pointed to the correct location. After the child pointed to the correct corner, he was
instructed to turn himself ®ve full rotations with his eyes closed. The experimenter
asked him to switch directions once during these ®ve turns and then had him stop
facing a randomly determined wall, so that two of the corners were immediately
visible. He was asked to open his eyes, `look around the room', and retrieve the
object. The object's hiding location and the child's facing position and ®rst search
were recorded by the experimenter and by the video camera. If the child failed to
®nd the object after two search attempts, the experimenter told him where it was.
Once the child had located the object, a new hiding place was chosen and the next
search trial commenced, until a total of four to six trials had been conducted,
depending on the child's level of interest in the game. After the conclusion of the
trials with the red wall landmark, the experimenter removed the red fabric in view of
the child and told him that a few more trials would be run under these new
conditions. Trials in the all-white chamber proceeded as before except that a
minimum of two trials and a maximum of ®ve trials were given, with most
subjects completing three or four trials. For each task, the main dependent
measure was the proportion of search at the correct corner for each condition.
126.96.36.199. Nonverbal intelligence. The Test of Nonverbal Intelligence-II (TONI-II),
made by Pro-Ed Corporation (Austin, TX), was used. Children were administered
®ve training items and once they had understood the procedure, they took the test in
the standard order (with problems running from easiest to hardest). Scores were
computed as percentiles and recorded on the Pro-Ed forms.
188.8.131.52. Language production task. This task was modeled on the production
assessment task used by Cox and Richardson (1985). Two three-dimensional
grids of plexiglass panels, each 30.5 £30.5 cm and arranged `on top' of each
other spaced 20.3 cm apart (held in place by four 1.3 cm dowel rods per each set
of three panels), were used. The middle layer of the grid contained a central 2.5 cm
hole surrounded by four other 2.5 cm holes spaced 3.8 cm away from the center hole
in a cross-like pattern (so that one hole was directly in front of the center hole,
another was directly behind it, and the two other holes were directly to the center
hole's left and right). The top and bottom layers of the grid contained a single central
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 277
2.5 cm hole, so that there was one hole directly above the middle central hole and
another 2.5 cm hole directly below the middle central hole. A speckled red, green,
yellow and white ping pong ball served as a central reference point for the two
language tasks and was located in the middle central hole. Subjects had three Koosh
balls (balls approximately 8 cm in diameter with a rubber core and pieces of colored
rubber emanating from that core), including a blue and purple one, a green and blue
one, and a red one with a face and legs on it, to use to complete the language tasks.
For the language production task, the child sat next to the parent, both facing the
same allocentric direction and each facing a plexiglass grid of their own. The
experimenter showed the child the Koosh balls which the experimenter would be
placing on his grid, and the experimenter also showed him three identical Koosh
balls which his parent would be allowed to place on her grid. The child was told that
the experimenter would place the Koosh balls in different positions around the
central red speckled ball and showed the child the six possible positions for a
Koosh on the grid without using any language to describe them. The
experimenter told the child that the child's job was to describe where the Koosh
ball was `in terms of the red speckled ball' so that his parent could make her grid and
Kooshes look just like the child's. Finally, the child was told that the task would be
too easy if the parent could simply look at her child's grid to see what she should do
and that a large barrier would be placed between the two grids so that the parent
couldn't see what to do, but would have to rely on her child's descriptions for
guidance. The task commenced with the experimenter placing the blue Koosh
ball under the red speckled ball, and the child had to describe this arrangement to
the parent. Sixteen trials were run in total, with four testing the child's ability to
produce phrases with `above' and `below' (or `on top of' and `under' or any other
equivalent terms), four testing the ability to produce `in front of' and `behind', four
testing `to the left of' or `on the left side of' and four testing `to the right of' or `on
the right side of'. Trials were given in the same order to each subjects and were a
random assortment of the types of phrase. The eight `to the left/right of' trials tested
the child's linguistic knowledge of the type of information needed to solve the
landmark-based reorientation task. An example of such a phrase that the child
needed to produce was: `Put the blue Koosh to the left of the red ball.' The three
types of phrase were tested to get a general measure of the child's spatial language
ability. Children's scores were computed as the number of correct `left'/`right'
production trials (out of eight) and as the number of correct `other production'
trials (`above/below' and `in front/behind', out of eight), and recorded as
percentages correct for each category of spatial term production. In a few cases,
children used exactly the opposite of the correct L or R terms on 100% of LR trials
(e.g. saying `on the left side' for the right side and vice versa), but these responses
were scored as incorrect. Additionally, in two cases, two children gave `correct' LR
trials by using egocentric information from their body rather than LR terms (e.g. `put
the blue Koosh on the side of my [wrist] watch next to the red speckled ball' instead
of `on the L side of the red ball'). Because these cases fully speci®ed the information
needed to solve the reorientation task, they were counted as correct; their effect on
the results was minimal due to their small number.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299278
184.108.40.206. Language comprehension task. This task was also modeled on the tasks of
Cox and Richardson (1985), and it made use of one of the three-dimensional grids
from the language production task. The experimenter showed the child the grid
again and gave him the same three Koosh balls as were used in the previous task,
telling him: `This time I'm going to stand behind you [so that the two people's
reference frames were the same] and I'm going to tell you where to put your
Kooshes!' Sixteen trials like the trials before proceeded, except that the phrases
were reversed so that the ®rst one was: `Put your blue Koosh ball on top of the red
speckled ball.' Again, there were 16 trials in total, with four testing `above'/`below',
four testing `in front of'/`behind', and eight testing `to the L/R of'. Scores were
computed and recorded in the same manner as for the spatial language production
task (e.g. with exact reversal of LR terms counting as wrong).
220.127.116.11. Digit span task. This task was a simpli®ed version of the digit span tests that
typically appear on IQ tests. The child was told that he would hear the experimenter
`say some numbers' (with `numbers' meaning `sequences of digits') and that his job
would be to `say the numbers back right after you hear them'. There were two
sequences of digits of each length given, on two consecutive trials, starting with
single digit numbers and incrementing by one every two trials. Trials with more than
one digit did not proceed until the child had correctly responded on both single digit
trials, indicating understanding of the task. Each child was given the same sequences
of digits, which were read to him by the experimenter at a rate of approximately two
digits/s. The child was given two trials with sequences of each length to repeat back
to the experimenter. If he got both trials right, the experimenter progressed to the
next digit length. Additional shorter-sequence trials were interspersed among the
longer trials for subjects whose motivation seemed to be ¯agging. After a child
failed two sequences within a given length or from one length to the next in a
row, the test stopped and the child's digit span was considered to be the level just
prior to where he stopped getting the sequences right (e.g. if he got both ®ve-digit
sequences right but missed both six-digit sequences, his digit span was considered to
be 5, and if he got both ®ve-digit sequences right and one six-digit sequences right,
and made an error on the other six-digit sequences and on the seven-digit sequence,
his digit span was considered to be 5.5).
18.104.22.168. Serial visuospatial span task. This test was similar to other tests of
visuospatial memory span (e.g. Smyth & Scholey, 1996; Swanson, 1999). It made
use of a single-layer grid which consisted of a white foamcore board with
dimensions of 30.5 £30.5 cm with a 3 £3 array of green dots. On this test,
subjects saw the experimenter take a single Koosh ball and place it in a series of
positions on the grid. Their job was to `do exactly what [the experimenter] did' and
move the Koosh ball through the same series of positions. Two trials were given of
each sequence length, as with the digit span test, and in order of increasing length,
until the child had failed 50% of the trials at one level and all the trials at the next
level or had gone from no failures to 100% failures. As with the digit span test, the
®rst trial consisted of a single position only, and once the subject got two such trials
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 279
right the experimenter moved on to a sequence of two positions. The experimenter
moved at a rate of approximately two positions/s, as with the digit span test. The
results were scored as for the digit span test.
The analyses for this experiment were conducted in several stages. The ®rst stage
was to analyze the entire group of children's red-wall and white-wall reorientation
performance, to determine whether the ®nding from Experiment 1 that children in
this age range are beginning to show adult-like reorientation performance in the
room with an indirect non-geometric landmark was replicated, as well as to probe
for effects of the variable subject sex. Only subjects' absolutely correct search
percentages were analyzed for each room, because the room's corners were not
distinguishable from one another with geometry (since the current room was
square), and because two corners instead of one corner were proximal to the correct
corner in the white room. ANOVAs were run as in Experiment 1 to probe for effects
of condition and subject sex, with Type III statistics and results used as before.
Single-sample t-tests were used to test the speci®c hypotheses that children's search
would be above-chance in the red-walled room but at chance in the white-walled
room, and that red-wall search would exceed white-wall search.
The next stage of analysis concerned the multiple regressions. First, tables of
means, standard deviations, and ranges as well as histograms were constructed for
all measures with attention paid to the degree of variability shown by each one.
Next, a simple correlation matrix was computed for all measures in order to further
assess their predictiveness, for possible inclusion in the multiple regression model.
Only measures that showed substantial variation (e.g. subjects' scores varied by over
50% or more of the possible range), were not at a mean of chance, and correlated
with at least one other measure were included in this multiple regression model.
The multiple regression work was done in two substages. In the ®rst substage, all
potential regressors were included in the model attempting to predict red-wall reor-
ientation, including those that were at chance or which did not correlate with any
other measure in the simple regression. Although those measures were at that time
unlikely to be included in the ®nal regression analysis, this substage provided the
®nal evidence that those variables did not improve the model and could safely be
excluded. In the second substage, the smaller and ®nal group of regressors were
analyzed to test their ability to predict red-wall reorientation performance given the
contributions of all the other variables, using SAS version 6.09 for mainframe's
PROC REG. The red-wall data for children scoring 0 on any measure found to
correlate signi®cantly with red-wall reorientation were also tested for being signi®-
cantly above chance, to examine further whether any non-signi®cant variables were
contributing in even a minor way to the development of adult-like reorientation
Finally, a series of post-hoc analyses were conducted to test further hypotheses
that arose from the initial results. These analyses relied on t-tests and were done
using subsets of the total dataset, as described in Section 3.2. The results of these
analyses were subjected to post-hoc Bonferroni corrections (Darlington, 1990).
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299280
Fig. 5 presents the entire sample's search in the all-white room and the red-walled
room. It can be seen from inspection that the search in the white-walled room was
distributed at chance among the four corners and that the search in the red-walled
room tended to be concentrated in the correct corner although there was substantial
variability. An ANOVA on the subjects' percent of search absolutely correct in the
red-walled room versus in the white-walled room, with the within-subjects factor of
condition and the between-subjects factor of sex, con®rmed this impression. The
search in the two conditions differed (F1;2213:56, P0:0013), re¯ecting the
subjects' tendency toward correct search in the room with one red wall. In contrast,
the sexes did not differ in their search patterns either overall (F1;22,1:65,
P.0:60) or in either condition (F1;22,1, P.0:70). Single-sample t-tests
con®rmed and extended this general pattern of results. The children's search at
the correct corner in the white-walled room was at chance level (jt23j ,1,
P.0:40), whereas their search at the correct corner in the room with one red
wall exceeded chance level (t235:053, P,0:0001). Their red-wall reorienta-
tion exceeded their white-wall reorientation (t234:004, P0:0006).
A further ANOVA was conducted on the data from all subjects, and focused on
their trial-by-trial performance. The analysis with the within-subjects factor of trial
(four, using the ®rst four trials given by each subject since all subjects gave at least
four good trials in the red-walled room) revealed no trial effect (F3;69,1,
In preparation for the multiple regression analysis, each measure was assessed
separately to evaluate its appropriateness for inclusion in the regression model.
Means, standard deviations, and ranges (Table 1) as well as histograms for each
measure indicated that there was substantial variation on each measure. A simple
correlation matrix among all measures (data not shown) revealed that all measures
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 281
Fig. 5. Search patterns of children in Experiment 2. (a) In the all-white room. (b) In the room with one red
but two had enough variation and predictiveness to exhibit a trend toward correlat-
ing with at least one other measure (P,0:25). The two measures failing to meet
this standard were white-wall reorientation scores, which were at chance as ascer-
tained above, and sex, whose simple correlations with other measures all had P
values of greater than 0.50. Interestingly, production of phrases involving `left' and
`right' (`LR production') was highly signi®cantly and positively correlated with red-
wall reorientation performance (r0:612, P0:002; Fig. 6), with better produc-
tion performance correlated with better red-wall reorientation. Age was marginally
positively correlated with performance in the red-walled task (r0:38,
P0:0694), with older children performing better than younger ones.
A `test' model was run including the independent measures sex and white-wall
reorientation along with other independent variables to make sure they could be
safely deleted from the ®nal model. In accord with prediction, neither variable
accounted for even close to a signi®cant amount of variation in red-wall perfor-
mance with the contributions of other variables taken into account (P.0:55). A
subsequent model was therefore run without either of those variables.
The ®nal model used red-wall reorientation as the dependent variable and age, IQ,
digit span, spatial span, comprehension of other spatial phrases (`other comprehen-
sion'), LR comprehension, other production, and LR production as independent
variables. The overall model was marginally signi®cant (F8;152:483,
P0:0613). Of the individual factors, only LR production was signi®cant with
other variables accounted for (Table 2, P0:0359). Other spatial preposition
comprehension showed a trend toward being negatively correlated with red-wall
reorientation with other variables accounted for (P.0:10).
In a further, planned analysis, the subset of children scoring 0 on the measure of
production of phrases involving right and left was analyzed separately to determine
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299282
Descriptive statistics for each measure in Experiment 2
NMean Standard deviation Minimum Maximum
RW 24 0.552 0.298 0.00 1.00
WW 24 0.285 0.202 0.00 0.670
SEX 24 1.58 0.504 1.00 2.00
AGE 24 5.80 0.206 5.40 6.20
IQ 24 0.847 0.142 0.420 0.980
DIG 24 5.48 1.12 4.00 8.00
VIS 24 3.58 0.803 2.00 5.00
COT 24 0.734 0.279 0.00 1.00
CLR 24 0.731 0.396 0.00 1.00
POT 24 0.479 0.292 0.00 1.00
PLR 24 0.445 0.401 0.00 1.00
RW, red-wall reorientation; WW, white-wall reorientation; SEX: 1, male; 2, female; DIG, digit span
test; VIS, visuospatial span test; COT, other spatial term comprehension; CLR, comprehension of phrases
involving `left' and `right'; POT, other spatial term production; PLR, production of phrases involving
`left' and `right'.
whether their reorientation was at chance level. Eight of the 24 subjects scored 0 on
this language test and did not search above the chance rate at the correct corner in the
reorientation task (25%) (jt7j ,1:5). In contrast, subjects receiving the eight
lowest scores on the IQ test searched at the correct corner signi®cantly above chance
even though their group size was identically small (t73:371, P0:0019).
A series of post-hoc analyses further explored the relationship between LR
production and other language measures and red-wall reorientation performance.
A total of ten post-hoc tests were performed, leading to a Bonferroni correction
factor of 10 (Darlington, 1990). For these analyses and on the basis of their language
performance histograms, subjects were divided into four groups: (1) subjects scoring
low (0 or 1 out of 8) on LR production; (2) subjects scoring high (7 or 8) on LR
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 283
Fig. 6. Simple correlation between LR production and red-wall task performance in Experiment 2.
Output for each variable in ®nal multiple regression model for Experiment 2
Coef®cient Standard error tvalue Pvalue
INT 21.83 2.24 20.817 0.427
AGE 0.353 0.388 0.909 0.378
IQ 0.215 0.470 0.457 0.654
DIG 20.008 0.063 20.123 0.904
VIS 0.085 0.077 1.11 0.285
COT 20.394 0.230 21.72 0.107
CLR 0.089 0.144 0.617 0.547
POT 20.101 0.217 20.467 0.647
PLR 0.375 0.163 2.31 0.036
production; (3) subjects scoring low (0 or 1 out of 8) on LR comprehension; and (4)
subjects scoring high (7 or 8) on LR comprehension but low (0 or 1) on LR produc-
tion. Fig. 7 presents the search patterns in the red-wall task for these four groups.
The ®rst tests focused on the production groups. An unpaired t-test compared the
red-wall absolutely correct percentages between the low and high production groups
which together accounted for 16 of the 24 subjects. This test revealed that high LR
production subjects searched at the correct corner signi®cantly more than low LR
production subjects even after post-hoc correction (t1523:47, corrected
P0:034). The low production group searched the correct corner only at chance
rate (t7,1:25), whereas even after post-hoc correction the mean of the high
production group exceeded chance level (t54:868, corrected P0:046).
A similar analysis was done to test the hypothesis that LR comprehension might
be suf®cient to allow adult-like reorientation performance. For example, it was
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299284
Fig. 7. Red-wall search performance by subjects in the different language comprehension and production
groups of Experiment 2, as described in the text. (a) The low LR production group (N10). (b) The high
LR production group (N6). (c) The low LR comprehension group (N5). (d) The group with high LR
comprehension but low LR production (N5).
possible that LR production signi®cantly predicted red-wall reorientation whereas
LR comprehension did not because the ®rst measure showed more variability than
the second. A t-test compared the red-wall reorientation performance of the low
comprehension group with that of the group scoring high on comprehension but low
on production. Even in the uncorrected results, the two groups did not differ from
each other (jt8j ,1, P.0:45). Moreover, t-tests con®rmed that neither group
performed above chance (for the low comprehension group, jt4j ,2:10, uncor-
rected P.0:10; for the high comprehension but low production group, jt4j ,1,
uncorrected P.0:70). These results suggest that LR comprehension alone is not
suf®cient for ¯exible reorientation performance in the red-walled room.
Subjects receiving a high LR production score were also compared to subjects
receiving a high LR comprehension score but a low LR production score. A t-test
comparing these two groups suggested that subjects with high LR production ability
outperformed subjects with high LR comprehension but little or no LR production
ability (t93:42, corrected P0:075). T-tests showed that the former group
performed above chance in the red-walled room (t54:868, corrected
P0:046) but the latter group did not (jt4j ,1).
Finally, an ANOVA with the within-subjects factor of trial probed for the learning
effect across trials 1±4 in subjects who scored high on LR production, an effect
suggested by some but not all of the results of Experiment 1. This test indicated that
subjects' performance did not change over the course of these trials
(F3;15,1:20, uncorrected P.0:35).
The results from Experiment 2 con®rm and extend the ®ndings from Experiment 1
with older children in an Indirect Landmark task. In Experiment 2, subjects of
approximately 6 years of age were again found to make signi®cant use of the
non-geometric cue in the red-walled Indirect Landmark task, whereas they
performed at chance in the all-white room. However, as in Experiment 1 not all
subjects displayed this pattern of results. Instead, some subjects performed at chance
in the room with the red wall, some performed well above chance with a few
correctly locating the object on every trial, and some were in the middle.
With both the simple correlation matrix and the multiple regression, a single
factor was found to predict red-wall task performance: production of phrases invol-
ving `left' and `right'. The LR production test assessed children's ability to conjoin
object property information with sense (`left' or `right') information to describe a
visual scene containing features with these dimensions (e.g. `put the blue Koosh ball
to the left of the red ball'). This requirement was similar to the propositional
demands of correctly coding the hidden object location in the room with one red
wall (e.g. `the toy is in the corner to the left of the red wall'). Subjects who scored
poorly on the LR production test tended to perform poorly on the red-wall task, and
subjects who scored well on the production task in general scored well on the red-
A series of statistical tests explored whether it was LR language production per se
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 285
that predicted red-wall performance or whether knowing one's left and right or
comprehending phrases involving left and right was suf®cient. In one of these
explorations, subjects who scored high on the LR comprehension test but poorly
on the LR production test were compared with subjects who scored poorly on the LR
comprehension test. Subjects who scored high on LR comprehension but poorly on
LR production failed to perform better in the red-wall task than subjects scoring
poorly on the LR comprehension task. Moreover, each group performed at chance in
the room with one red wall. Since subjects scoring high on the comprehension test
presumably knew their left and right sides (consistent with the age at which this
ability typically develops; Cole & Cole, 1996) and had conceptual knowledge invol-
ving left and right, these factors appear insuf®cient to allow children to overcome the
phylogenetically older, geometric reorientation capacity and solve this task by using
more types of information.
In contrast, the group scoring high on the LR production test performed better in
the room with one red wall than did subjects scoring high on LR comprehension but
low on LR production. Furthermore, this group performed above chance in the red-
wall task, unlike the high comprehension, low production group. These ®ndings
strongly suggest that whereas the ability to produce phrases involving `left' and
`right' is associated with success in the red-wall task, successful comprehension of
phrases involving these terms without the ability to generate such phrases oneself is
not associated with red-wall success.
It could be argued that any of several potential problems with the dataset
produced only a single signi®cant correlation with red-wall performance. For
example, if there was more variation in the LR production measure than in the
LR comprehension measure, LR production might spuriously have seemed to be
the only signi®cant predictor of red-wall performance when in reality, LR compre-
hension alone might be suf®cient. However, the correlation matrix showed that LR
comprehension exhibited enough variation to correlate with at least one other
measure, arguing against this being the reason for a lack of a correlation with
red-wall performance. In fact, all measures included in the ®nal multiple regression
model correlated with at least one other measure in the simple correlation matrix,
supporting their validity and predictive power. These facts suggest that either LR
production alone, or LR production in combination with LR comprehension (since
nearly all subjects scoring high on LR production also scored high on LR compre-
hension) is closely tied to locating the object in the red-wall task by using an
indirect non-geometric cue.
In contrast, a number of alternative accounts for some children's advanced perfor-
mance were not supported by these ®ndings. For instance, general increases in
cognitive capacity, such as the gains in verbal and spatial working memory that
come with age through early adulthood (Baddeley, 1990; Kuhn, 1992; Swanson,
1999) which should theoretically allow children to detect more correlations (e.g.
Halford, Wilson, & Phillips, 1998), were not linked to reorientation task perfor-
mance in Experiment 2. Intelligence and increased world knowledge also did not
appear to cause those children's advances, as indicated by the lack of effects in our
experiments with IQ and age. The onset of hippocampal maturity, which occurs at
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299286
approximately 3.5 postnatal years in humans (Rudy, 1991), also does not provide a
convincing account, as suggested by the fact that children in the age range of 3.0±4.0
years showed no ability to solve the Indirect Landmark task, and the fact that a large
subset of children in the 5±7 year age range continued to solve that task like young
children. Finally and again, neither simply knowing one's left and right sides
(knowledge that children in this age range are also acquiring; Cole & Cole,
1996), nor having the conceptual knowledge suf®cient for comprehending spatial
prepositional phrases containing these terms explains the change in reorientation
abilities, for only LR production, not LR comprehension, correlated with advanced
This experiment thus revealed a speci®c correlation between the ability to
produce phrases involving `left' and `right' and the ability to overcome the phylo-
genetically more common, geometry-based reorientation process to use construc-
tions involving spatial, left/right information and non-spatial wall color information.
Assuming for the moment that the relationship is causal (we present evidence that it
is in Section 5), the bene®t conferred by this linguistic ability could take any of a
variety of forms. For example, it might only serve the purpose of extending the
abilities in the domain of reorientation by allowing conjunction of information not
normally used to solve reorientation tasks, with a bene®t speci®c to that situation.
Another possibility is that it serves the more general purpose of bringing information
from disparate domains of reasoning together to solve a variety of tasks in evolu-
tionarily novel ways, allowing the limits of other phylogenetically older processes to
be transcended as well. Note that even if this latter possibility were correct, language
might in addition have other effects on cognitive processes.
In the third experiment, we investigated these possibilities by testing new subjects
in the prior language production tasks and by testing their ability to solve a different
spatial memory task, which also required the conjunction of spatial and non-spatial
information. This task was modeled on a task run with adult rats by Biegler and
Morris (1993, 1996). In this task, food was hidden in a constant spatial relationship
in relation to a movable array of landmarks. On each trial the landmarks, and there-
fore also the food, were moved to new locations, remaining in the same relationship
to one another but changing in their relationship to ®xed features of the environment.
Rats rapidly learned to concentrate their search for the food in the general area of the
landmark array, but required hundreds of trials to learn to combine landmark and
precise spatial information to search in the exact food location. In a similar task, and
in contrast to adult rats, adult humans were found to search in the correct location
even on their ®rst trial (Hermer-Vazquez et al., 1999).
Experiment 3 tested the spatial phrase production and moving object search
abilities of older children. If the correlation between LR production and search
ability held in this new task, it would suggest that language plays a more general
role in rapidly combining different types of information to solve a variety of tasks in
novel ways. If not, it would suggest that the bene®ts of language for spatial memory
operate only in very restricted circumstances, perhaps even only for conjoining
information across the domains of reorientation by environmental shape and remem-
bering non-spatial object features.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 287
4. Experiment 3
In this study, a new group of older children was tested on a version of the Biegler
and Morris (1993, 1996) task adapted for children (Fig. 8). Subjects stood at the
center of a circle of nine plastic cups and were shown a large, fronted landmark
object standing in between two of the cups. The child watched as an object was
hidden in the cup to the landmark's right. Then the child's eyes and ears were
covered and the landmark and toy were moved to new positions such that the toy
again was hidden in the cup directly to the right of the landmark. The child's eyes
and ears were uncovered and he was asked where the toy was now. After giving an
answer, the child was shown where the hidden toy was and seven to nine further
trials were given.
In our task, we expected that children would show an ability similar to that of
rats to con®ne their search to the two cups nearest the landmark. However, it was
an open question whether any children would exhibit a search pattern unseen in
adult rats until after hundreds of training trials: a tendency to locate the hidden toy
in the cup to the right of the dwarf, requiring a combination of landmark and sense
information. After three demonstration trials, the percentage of trials on which
each subject located the toy in one of the two cups next to the dwarf, as well as
the percentage of trials on which he located it in the cup directly to the dwarf's
right, were tallied.
The children's performance on this task was analyzed in relation to their age, sex,
LR production abilities, and abilities to produce other spatial phrases, all as
measured in Experiment 2.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299288
Fig. 8. The environments for the movable object search task given to children in Experiment 3. (a) The
appearance of the room after hiding of the toy. The hidden object was under the cup to the landmark's
right and was not visible. (b) The appearance of the room for search, after the landmark and toy had been
moved to new locations.
Seven boys and ®ve girls between the ages of 6.2 and 7.2 years (mean age 6.6
years) came in for the one-visit experiment, and were recruited and paid as before.
All the children provided a complete set of usable data.
For the moving object search task the same square experiment room with one red
wall was used with additional objects inside it (Fig. 8). The plastic dwarf that
children saw outside the experiment room before they entered and then used as
the object around which space would be represented was 122 cm tall by approxi-
mately 23 cm on average in diameter, with a bright blue suit and with gold hair, a
blue cap and red boots. The cups in which a small toy could be hidden were opaque,
red, plastic, party cups, 11.4 cm high by 9.5 cm in diameter at the drinking end. For
the language production task, the same materials were used as in the previous study.
All subjects received the eight to ten moving object search trials, with initial and
®nal hiding positions randomized. The ®rst three trials were considered demonstra-
tion trials and only trials following these ®rst three were analyzed. (During the ®rst
three trials, subjects learned not to look in the old, allocentric location of the toy.)
After completing the moving object test, subjects gave 16 language production trials
as in Experiment 2.
Before entering the testing room, subjects were `introduced' to the movable land-
mark, the large plastic dwarf, outside the testing room, and told, `He likes to play a
game in which you and he go into a room together, a toy is hidden, your eyes and
ears are covered, and then you have to ®nd the toy.' The subject then went to the
center of the testing room with the experimenter, who said, `Watch what I'm doing',
and placed the dwarf in between two of the cups and placed the toy under the cup to
the dwarf's right. The experimenter covered the child's ears with earmuffs and eyes
with a blindfold which also held the earmuffs in place tightly, and then moved the
dwarf and toy to new positions, preserving the relationship between them. The
experimenter said to the child, `Where's the toy? Go get it', and recorded the
subject's subsequent search. If the child failed to locate the toy, the experimenter
retrieved it while the child watched. Seven to nine more trials were given in the same
4.1.5. Coding and analyses
Only the search after the ®rst three trials was coded and analyzed. The search was
coded in two ways: whether it was in the correct location relative to the dwarf (i.e. on
its right side) and whether it was in either of the two spots next to the dwarf. Coding
of the language task took place as before. The data were analyzed in a manner
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 289
similar to that used in Experiment 2. Initial ANOVAs were run on the two dependent
measures to check for an effect of sex on the data. Descriptive statistics were
compiled on the data and histograms were studied to determine whether each vari-
able had suf®cient variability to be included in the multiple regression models. After
the multiple regressions were completed, planned t-tests were run to test speci®c
hypotheses about relationships in the data.
Table 3 shows the means, standard deviations and ranges for all the measures in
this experiment. Children exhibited a range of scores on the moving object propor-
tion correct measure, but little range of scores on the moving object proportion next
to the dwarf, on which subjects uniformly scored well. Fig. 9 shows the search
distributions in the moving object task for all subjects. This group of subjects did
not search randomly for the hidden object, searching in the correct location more
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299290
Descriptive statistics for each measure in Experiment 3
NMean Standard deviation Minimum Maximum
NEXT 12 0.960 0.073 0.830 1.00
CORR 12 0.718 0.249 0.290 1.00
SEX 12 1.42 0.515 1.00 2.00
AGE 12 6.56 0.255 6.20 7.20
POT 12 0.573 0.135 0.380 0.750
PLR 12 0.543 0.450 0.00 1.00
NEXT, proportion of search on either side of landmark; CORR, proportion of search in the correct
location to the landmark's right.
Fig. 9. Search results for the whole group of children in Experiment 3. E, search that was neither
proximally nor absolutely correct.
often than by chance (t118:489, P,0:0001), and they searched in the two
locations next to the dwarf far more often than expected by chance (t1135:242,
P,0:0001). It can readily be seen that on nearly all trials, subjects searched in one
of the two cups adjacent to the dwarf's new position. Additionally, as a group
subjects searched more often in the cup to the dwarf's right, the correct position,
than in the cup to the left of the dwarf (t113:529, P0:0047), although there
was substantial variation among subjects in this tendency (Table 3). Initial one-way
ANOVAs with the factor sex revealed that the sexes did not differ in their tendency
to search in the two cups to the right and left of the dwarf (F1;10,1, P.0:40),
and neither sex tended to search the absolutely correct location more often
The correlation matrix for independent measures and subjects' absolutely correct
search in this task revealed that all variables showed at least a trend (P,0:15)
toward correlating with at least one other measure (data not shown). This fact further
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 291
Fig. 10. Simple correlation between LR production and moving object search in Experiment 3.
Output for each variable in ®nal multiple regression model for Experiment 3
Coef®cient Standard error tvalue Pvalue
INT 20.632 1.76 20.358 0.731
SEX 20.082 0.146 20.560 0.593
AGE 0.188 0.296 0.635 0.545
POT 0.098 0.649 0.151 0.884
PLR 0.319 0.154 2.08 0.077
supported the appearance from the means and standard deviations that these
measures showed substantial variability. All these variables were therefore retained
for the regression analysis. It should also be noted that one independent measure
correlated signi®cantly with absolutely correct moving object search performance,
namely LR production (r0:68, P0:0153; Fig. 10).
The resulting regression model attempted to predict absolutely correct moving
object search scores from the independent variables sex, age, LR production, and
other production. The overall model showed a trend toward signi®cance
(F4;72:053, P0:1910). The only independent measure to approach a signif-
icant correlation with correct moving object search, with other variables controlled
for, was LR production (t72:076, P0:0766; Table 4).
The y-intercept of the graph of moving object correct search by LR production
score (Fig. 10), showing the average moving object score of the three subjects
scoring 0 on the LR production task, is close to the expected value of 0.50. Although
they, like other subjects, tended to concentrate their search in the two locations
proximal to the dwarf (observed proportion 0:88, chance expectation 0:22,
t23:381, P0:0774), they did not search in the correct location more than
the location to the other side of the landmark (observed proportion 0:52, chance
expectation 0:50, jt2j ,1). However, neither the three subjects scoring the
lowest on other production, nor the three youngest subjects searched that location
signi®cantly above the 50% level either, although their mean correct search scores
were slightly higher than those of subjects scoring 0 on LR production (t2,1:10;
in both cases, there was substantial overlap with the low LR production group). This
shows the problems of conducting the analyses on only three subjects.
Further planned t-tests examined the relationship between low and high LR produc-
tion groups as de®ned in Experiment 2 and moving object search. Fig. 11 presents the
distribution of search for these two production groups in the moving object search
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299292
Fig. 11. Search patterns for the different LR production groups in Experiment 3. (a) Search by the low LR
production group (N4). (b) Search by the high LR production group (N6).
task. Subjects scoring a 0 or a 1 on the LR production task (N4) split their search
evenly between the correct location and the spot on the adjacent side of the dwarf. In
contrast, subjects scoring a 7 or an 8 on LR production (N6) searched in the correct
location 87% of the time and in the adjacent location only 11% of the time. Subjects
scoring high on the LR production test outperformed subjects scoring low on this test
on the moving object search task (t82:899, P0:0199). Subjects who scored
high on LR production also searched the absolutely correct cup more often than the
other proximal cup (t55:326, P0:0034), while subjects who scored low on LR
production showed no tendency to search under the correct cup more than the other
proximal cup (t3,1).
In the moving object search task, all subjects quickly learned to con®ne their
search to locations next to the landmark. In this regard they resembled the adult rats
tested by Biegler and Morris (1993, 1996), who learned to search for hidden food in
the region proximal to a movable landmark array. Children's success at this aspect
of the task suggests that by this age they have a mechanism that allows the associa-
tion of a reward such as obtaining a toy with a proximal and movable cue, in this
case the dwarf. This ®nding is consistent with what is known about adult rats'
mechanisms for forming such direct associations to solve many tasks in their envir-
onment, even when the cue that predicts reward does not occupy a stable, allocentric
position (Biegler & Morris, 1993; McDonald & White, 1993).
In contrast to subjects' proximal search, which showed little variation across
subjects, there was wide variation in subjects' ability to search the absolutely correct
location to the dwarf's right. Some subjects succeeded in combining landmark
information (relating to the dwarf) with precise spatial information to solve the
task on each trial. Subjects' tendency to search the absolutely correct location
showed a signi®cant simple correlation with LR production ability. In the multiple
regression model, controlling for the contribution of other variables, correct moving
object search marginally correlated with the subjects' LR production ability, similar
to the ®ndings in Experiment 2 (in which a signi®cant correlation was found in the
multiple regression). Moreover, subjects scoring high on LR production outper-
formed low LR scoring subjects on the moving object search task and scored
signi®cantly above the proximal chance rate of 50%, unlike low production subjects.
These ®ndings suggest that the conjunctive powers of language allow subjects to
combine information across the domains of object property representation and
space, to aid in the solution of this task as well as the reorientation task.
However, conclusions regarding language and moving object search from this
experiment must be considered preliminary because two important ®ndings were not
quite signi®cant. First, the correlation (with other factors accounted for) between
absolutely correct search and LR production was only marginally signi®cant. It is
possible that a larger sample size would have borne out the correlation in full (this
experiment used only 12 subjects, compared to 24 in Experiment 2). However, it is
also possible that the apparent correlation between LR production and absolutely
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 293
correct movable object search was spurious, and this issue will require further
studies. Second, the sample of subjects scoring 0 on LR production was small
(N3) and other samples of three subjects (those scoring lowest on other produc-
tion and those of the lowest ages), like the group scoring 0 on LR production, failed
to score above the chance level of 50% in their search. This raises the possibility that
even subjects without the language production capabilities in question may have
some ability to search the absolutely correct location above chance, but that tests on
only three subjects would not show it. In Section 5 we will present converging
evidence from studies with human adults suggesting that LR production does indeed
play a role, and in fact a causal role, in allowing the rapidly learned absolutely
correct moving object search displayed by normal human adults.
5. General discussion
These experiments reveal that the transition to more rapidly ¯exible performance
in these reorientation and movable object search tasks occurs fairly late in devel-
opment, before which time children exhibit the phylogenetically older spatial
memory capacities shown by adult rats.
Moreover, in both cases the developmental
transition to more human adult-like performance is associated with the onset of
language production capacities allowing the child to form linguistic representations
of the information needed to solve the tasks more ¯exibly. (Other factors such as
hippocampal maturity and minimal levels of intelligence and working memory
capacity may be necessary as well, but alone they are not suf®cient.) The current
®ndings, however, are merely correlational, leaving open multiple possibilities for
the relationship between developmental changes in spatial language production and
advances in spatial memory task performance. What is the nature of their relation-
Findings by Hermer-Vazquez et al. (1999) provide evidence that spatial
language production abilities allow human adult-like performance of the two
spatial memory tasks. They probed the reorientation and moving object search
abilities of human adults under a variety of interference conditions. In one of
these experiments, adults whose language production abilities were blocked by
performance of a concurrent verbal interference task (`shadowing', e.g. Posner,
1973) failed to reorient using a blue wall in a rectangular room, instead relying
only on the room's geometry. Their search was evenly split between the two
diagonally opposite, congruent corners of the room despite the fact that nearly
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299294
We do not suggest that younger children and rats never solve reorientation tasks by using non-
geometric information and never use exact spatial relationships in solving moving object search problems.
Indeed, and as mentioned above, under aversive conditions adult rats reorient themselves using non-
geometric cues (Dudchenko et al., 1997). Additionally, it has been reported that under some circum-
stances young children also solve reorientation tasks by using non-geometric cues (Learmonth et al.,
1998; N. Newcombe, pers. commun.). Finally, with extensive training adult rats come to solve moving
object search tasks as did the older children in Experiment 3 who succeeded (Biegler & Morris, 1996). But
under many circumstances young children, like adult rats, show surprising limitations whose transcen-
dence we explore in this paper and in Hermer-Vazquez et al. (1999).
all subjects reported after the experiment that they had noticed the blue wall. In
contrast, adults performing a nonverbal interference task of similar dif®culty to the
verbal interference task were able to use information about the blue wall as well as
in the reorientation task, performing in an adult-like manner despite the dual-task
scenario. Finally, verbally shadowing adults given various versions of the moving
object search task were able to rely on spatial information or non-spatial landmark
information, but not both, whereas non-shadowing adults easily conjoined these
types of information to solve the tasks. Taken together, these results strongly
suggest that the conjunctive powers of language production allow more ¯exible
performance of these tasks in humans. The current results are consistent with this
conclusion because not only would one expect a correlation if it were correct,
which we found, but one would also expect children scoring low on LR production
not to show any emergent tendency to search the correct locations above the
appropriate chance levels, which we also found.
Moreover, other evidence suggests that these language production skills may
have bene®ted our ancestors by allowing them to arrive at more ¯exible solutions
more quickly than the phylogenetically older processing systems permitted. Rats
tested in Biegler and Morris' moving object search task eventually learned to use
the movable landmark array to locate the hidden food, but it took them hundreds of
trials (Biegler & Morris, 1996). Likewise, in Experiment 3 children without LR
production skills failed to learn to use the movable landmark to locate the hidden
toy over the course of up to 12 trials. In contrast, children with those language
skills learned to solve the task more ¯exibly in little more than the initial three
training trials. Rats and children clearly differ cognitively in many other respects
such as general intelligence, and young children without LR production skills may
not require hundreds of trials to learn the moving object search task or conjunctive
reorientation task. However, the results of Experiments 2 and 3 suggest that
language production may allow older children to solve those tasks in the adult-
like manner more rapidly, by making some solutions more consciously accessible
to them more quickly (Block, 1996; Rozin, 1976). The biases inherent in a given
child's language, in how that language describes space and what its argument
structure is, may determine which solutions are rapidly available to the child versus
which ones will require many trials to learn.
It is possible that such a production-based mechanism, allowing the rapid combi-
nation of information from previously more separate processing systems, could
augment processing outside the realm of spatial memory as well. Although we
know of no evidence bearing directly on this point, Xu and Carey (1996) have
®ndings which suggest a similar role for language in bringing information from
disparate domains together. Their work suggests that some capacity related to
knowing the names of objects may allow two types of information about
objects, spatiotemporal information and object feature information, to be
merged into a more sophisticated object concept in humans (see Xu, 1999,
for discussion of how this might work). Simons (1996) provides converging
evidence for a role for language in object representation in his studies of human
However, we do not argue that these are the only ways in which language
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 295
augments cognitive capacities; indeed, many other effects of language have been
We also do not argue that it is impossible to solve the above tasks rapidly and
¯exibly without language. For the cases of spatial reorientation and memory for
object identity, at least some infrahuman species appear to have become able to
solve the tasks without linguistic representation. For example, Miller, Gouteux,
DeIpolyi, Santos, and Hauser (2000) have found that cotton-top tamarins solve
the Cheng reorientation task in a manner similar to that of human adults, by rapidly
conjoining geometric and non-geometric information. These ®ndings contrast with
those of Hermer-Vazquez et al. (1999), who found dependence on language in their
reorientation tasks with human adults. Similarly, while Simons (1996) reports that
human adults appear to require linguistic representation for memory of object iden-
tity, at least under some circumstances, rats as well as non-human primates routinely
solve delayed match and non-match to sample tasks with object property informa-
tion (e.g. Raffaele & Olton, 1988; Suzuki, Miller & Desimone, 1997). When linguis-
tic representations are required for a given task and species and when they are not
will clearly require further study.
However, we believe that the mechanism we have uncovered, which we have
found under multiple testing circumstances with children and with human adults,
may allow humans to solve a variety of tasks for which there was no prior specia-
lization. This mechanism may provide part of the answer to the question of how so
many cognitive and social changes were possible for the emerging human species in
such a short evolutionary time. The current line of research also suggests that a form
of `cognitive archeology' (Mithen, 1996) which probes the development of extant
species as well as their adult cognitive organization may be useful in understanding
that species' evolutionary history.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299296
Whether humans require language to conjoin spatial and non-spatial information is clearly task-
dependent. For example, we found that language production abilities were not required for children to
solve an oriented conjunction memory task, in which they had to remember which colored object was
located where on a 3 £3 grid (Hermer-Vazquez, manuscript in preparation). Moreover, in a reorientation
task in which different objects were hidden in different locations in a rectangular room, 3±4-year-olds
were able to remember which object was hidden in each location, using a combination of object identity
and geometric information, although they did not solve the reorientation task using the blue wall that was
present (Hermer, 1997b). Further research will be needed to resolve when language is required and when
it is not.
There are myriad other examples of the ways in which language augments or otherwise changes
cognition. For instance, DeVilliers and deVilliers (in press) present compelling evidence that children's
understanding of false beliefs is facilitated by acquiring the syntax of complementization. Language has
also been shown to alter cognition by strengthening episodic memory (Bauer & Wewerka, 1995), by
affecting the number of items that can be held in verbal working memory, thereby slightly facilitating
mathematical reasoning (Ellis & Hennelly, 1980; Hoosain & Salili, 1988; Naveh-Benjamin & Ayres,
1986), by altering one's primary coordinate system for spatial reasoning (Levinson, 1992), by distorting
long-term memory for spatial layout (Hirtle & Mascolo, 1986), by slightly facilitating counterfactual
reasoning (Au, 1983, 1984), and by providing a medium for rote-learned reasoning and by augmenting
more complex numerical reasoning in ways not yet understood (Dehaene, Spelke, Pinel, Stanescu, &
Tsivkin, 1999), to name only a few.
This study was supported by a predoctoral fellowship from NIH to L.H.-V. (1 F31
MH10607), by fellowships from the Cornell Cognitive Studies Program to L.H.-V.,
and by a grant to Elizabeth S. Spelke from NIH (R37 HD23103). We thank
Raymond Hermer-Vazquez and Elizabeth Spelke for discussion of the issues in
this manuscript, Elizabeth Spelke and Gail Ross for comments on the manuscript,
and Dick Darlington and Elizabeth Spelke for advice on the analyses.
Alberts, B., Bray, D., Lewis, J., Raff, M., Roberts, K., & Watson, J. D. (1992). Molecular biology of the
cell, New York: Garland.
Appenzeller, T. (1998). Art: evolution or revolution? Science,282, 1451±1454.
Au, T. K. (1983). Chinese and English counterfactuals: the Sapir-Whorf hypothesis revisited. Cognition,
Au, T. K. (1984). Counterfactuals: in reply to Alfred Bloom. Cognition,17, 289±302.
Baddeley, A. (1990). Human memory: theory and practice. Boston, MA: Allyn & Bacon.
Balter, M. (1998). Why settle down? The mystery of communities. Science,282, 1442±1445.
Bauer, P. J., & Wewerka, S. S. (1995). One- to two-year-olds' recall of events: the more expressed, the
more impressed. Journal of Experimental Child Psychology,59, 475±496.
Biegler, R., & Morris, R. G. M. (1993). Landmark stability is a prerequisite for spatial but not discrimina-
tion learning. Nature,361, 631±633.
Biegler, R., & Morris, R. G. M. (1996). Landmark stability: studies exploring whether the perceived
stability of the environment in¯uences spatial representation. Journal of Experimental Biology,199
Block, N. (1996). How can we ®nd the neural correlate of consciousness? Trends in Neuroscience,19,
Cheng, K. (1986). A purely geometric module in the rat's spatial representation. Cognition,23, 149±178.
Cole, M., & Cole, S. (1996). The development of children. New York: W.H. Freeman.
Darlington, D. (1990). Regression and linear models. New York: McGraw-Hill.
Dehaene, S., Spelke, E., Pinel, P., Stanescu, R., & Tsivkin, S. (1999). Sources of mathematical thinking:
behavioral and brain-imaging evidence. Science,284, 970±974.
DeVilliers, J. G., & deVilliers, P. A. (in press). Linguistic determinism and the understanding of false
beliefs. In P. Mitchell & K. Riggs (Eds), Children's reasoning and the mind.
Dudchenko, P. A., Goodridge, J. P., Seiterle, D. A., & Taube, J. S. (1997). Effects of repeated disorienta-
tion on the acquisition of spatial tasks in rats: dissociation between the appetitive radial arm maze and
the aversive water maze. Journal of Experimental Psychology: Animal Behavior Processes,23, 194±
Ellis, N. C., & Hennelly, R. A. (1980). A bilingual word-length effect: implications for intelligence testing
and the relative ease of mental calculation in Welsh and English. British Journal of Psychology,71,
Etienne, A. S., Maurer, R., & Seguinot, V. (1996). Path integration in mammals and its interaction with
visual landmarks. Journal of Experimental Biology,199, 201±209.
Gallistel, C. R. (1990). The organization of learning. Cambridge, MA: MIT Press.
Goodale, M. (1995). The cortical organization of visual perception and visuomotor control. In S. Kosslyn,
& D. Osherson (Eds.), An invitation to cognitive science (2nd ed.). Visual cognition (Vol. 2, pp. 167±
214). Cambridge, MA: MIT Press.
Gouteux, S., Thinus-Blanc, C., & Vauclair, J. (2000). Rhesus monkeys use geometric and nongeometric
information during a reorientation task. Manuscript submitted for publication.
Halford, G. S., Wilson, W. H., & Phillips, S. (1998). Processing capacity de®ned by relational complexity:
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 297
implications for comparative, developmental, and cognitive psychology. Behavioral and Brain
Hermer, L. (1994, March). The development and evolution of cognitive ¯exibility in humans: candidate
mechanisms. Poster presented at the 2nd annual meeting of the Cognitive Neuroscience Society, San
Hermer, L. (1997). The development and evolution of cognitive ¯exibility in humans: the case of two
spatial memory tasks. Published doctoral dissertation, Cornell University, Ithaca, NY.
Hermer, L. (1997b). Internally coherent spatial memories in a mammal. NeuroReport,8, 1743±1747.
Hermer, L. (1998, November). The development and evolution of cognitive ¯exibility in humans: the case
of two spatial memory tasks. Poster presented at the 28th annual meeting of the Society for
Neuroscience, Los Angeles, CA.
Hermer, L., & Spelke, E. S. (1994). A geometric process for spatial reorientation in young children.
Hermer, L., & Spelke, E. (1996). Modularity and development: the case of spatial reorientation. Cogni-
Hermer-Vazquez, L., Spelke, E. S., & Katsnelson, A. S. (1999). Sources of ¯exibility in human cognition:
dual-task studies of space and language. Cognitive Psychology,39, 3±36.
Hirtle, S. C., & Mascolo, M. F. (1986). Effect of semantic clustering on the memory of spatial locations.
Journal of Experimental Psychology: Learning, Memory and Cognition,12, 182±189.
Holden, C. (1998). No last word on language origins. Science,282, 1455±1458.
Hoosain, R., & Salili, F. (1988). Language differences, working memory, and mathematical ability. In M.
M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical aspects of memory: current research and
issues: Vol. 2. Clinical and educational implications (pp. 512±517). Chichester: Wiley.
Knierim, J. J., Kudrimoti, H. S., & McNaughton, B. L. (1995). Hippocampal place ®elds, the internal
compass, and the learning of landmark stability. Journal of Neuroscience,15, 1648±1659.
Kuhn, D. (1992). Cognitive development. In M. Bornstein & M. Lamb (Eds.), Developmental psychology:
an advanced textbook (3rd ed., pp. 211±271). Hillsdale, NJ: Erlbaum.
Learmonth, A., Newcombe, N., & Huttenlocher, J. (1998, November). Disoriented children use land-
marks as well as geometry to reorient. Paper presented at the annual meeting of the Psychonomics
Society, Dallas, TX.
Levinson, S. C. (1992). Language and cognition: the cognitive consequences of spatial description in
Guugu Yimithirr. Working paper #13, Cognitive Anthropology Research Group, Max Planck Institute
Margules, J., & Gallistel, C. R. (1988). Heading in the rat: determination by environmental shape. Animal
Learning and Behavior,16, 404±410.
McDonald, R. J., & White, N. (1993). A triple dissociation of memory systems: hippocampus, amygdala
and dorsal striatum. Behavioral Neuroscience,107, 3±22.
McNaughton, B. L., Barnes, C. A., Gerrard, J. L., Gothard, K., Jung, M. W., Knierim, J. J., Kudrimoti, H.,
Qin, Y., Skaggs, W. E., Suster, M., & Weaver, R. L. (1996). Deciphering the hippocampal polyglot:
the hippocampus as path integrator. Journal of Experimental Biology,199, 173±185.
Miller, C. T., Gouteux, S., DeIpolyi, A. R., Santos, L. R., & Hauser, M. D. (2000). Primate spatial
reorientation: coordinated use of geometric and landmark information following disorientation.
Manuscript in preparation.
Mithen, S. (1996). The prehistory of the mind: a search for the origins of art, science and religion.
London: Thames and Hudson.
Mittelstaedt, M. L., & Mittelstaedt, H. (1980). Homing by path integration in a mammal. Naturwis-
Naveh-Benjamin, M., & Ayres, T. J. (1986). Digit span, reading rate, and linguistic relativity. Quarterly
Journal of Experimental Psychology,38, 739±751.
O'Keefe, J., & Dostrovsky, J. (1971). The hippocampus as a spatial map. Preliminary evidence from unit
activity in the freely-moving rat. Brain Research,34, 171±175.
Posner, M. I. (1973). Cognition: an introduction. Glenview, IL: Scott Foresman.
Raffaele, K. C., & Olton, D. S. (1988). Hippocampal and amygdaloid involvement in working memoryfor
nonspatial stimuli. Behavioral Neuroscience,102, 349±355.
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299298
Ridley, M. (1993). Evolution. New York: Blackwell Scienti®c.
Rovee-Collier, C., Borza, M. A., Adler, S. A., & Boller, K. (1993). Infants' eyewitness testimony: effects
of postevent information on a prior memory representation. Memory and Cognition,21, 267±279.
Rozin, P. (1976). The evolution of intelligence and access to the cognitive unconscious. Progress in
Psychobiology and Physiological Psychology,6, 376±378.
Rudy, J. W. (1991). Elemental and con®gural associations. Developmental Psychobiology,24, 221±236.
Simons, D. (1996). In sight, out of mind: when object representations fail. Psychological Science,7, 301±
Smyth, M. M., & Scholey, K. A. (1996). The relationship between articulation time and memory perfor-
mance in verbal and visuospatial tasks. British Journal of Psychology,87, 179±191.
Suzuki, S., Augerinos, G., & Black, A. H. (1980). Stimulus control of spatial behavior on the eight-arm
maze in rats. Learning and Motivation,11, 1±18.
Suzuki, W. A., Miller, E. K., & Desimone, R. J. (1997). Object and place memory in the macaque
entorhinal cortex. Journal of Neurophysiology,78, 1062±1081.
Swanson, H. L. (1999). What develops in working memory? A life span perspective. Developmental
Tolman, E. C. (1948). Cognitive maps in rats and men. Psychological Review,55, 189±208.
Ungerleider, L. G., Courtney, S. M., & Haxby, J. V. (1997). A neural system for human visual working
memory. Proceedings of the National Academy of Sciences USA,95, 883±890.
Wang, R. F., Hermer, L., & Spelke, E. S. (1999). Mechanisms of reorientation and object localization by
children: a comparison with rats. Behavioral Neuroscience,113, 475±485.
Xu, F. (1999). Object individuation and object identity in infancy: the role of spatiotemporal information,
object property information, and language. Acta Psychologica,102, 113±136.
Xu, F., & Carey, S. (1996). Infants' metaphysics: the case of numerical identity. Cognitive Psychology,
L. Hermer-Vazquez et al. / Cognition 79 (2001) 263±299 299