Content uploaded by Bernhard Hommel
Author content
All content in this area was uploaded by Bernhard Hommel on Feb 17, 2015
Content may be subject to copyright.
Content uploaded by Bernhard Hommel
Author content
All content in this area was uploaded by Bernhard Hommel on Feb 17, 2015
Content may be subject to copyright.
ORIGINAL ARTICLE
Bernhard Hommel á Jo
È
rg Gehrke á Lothar Knuf
Hierarchical coding in the perception and memory of spatial layouts
Received: 7 October 1999 / Accepted: 17 February 2000
Abstract Two experiments were performed to investi-
gate the organization of spatial information in percep-
tion and memory. Participants were confronted with
map-like con®gurations of objects which were grouped
by color (Experiment 1) or shape (Experiment 2) so as to
induce cognitive clustering. Two tasks were adminis-
tered: speeded veri®cation of spatial relations between
objects and unspeeded estimation of the Euclidean dis-
tance between object pairs. In both experiments, veri®-
cation times, but not distance estimations, were aected
by group membership. Spatial relations of objects be-
longing to the same color or shape group were veri®ed
faster than those of objects from dierent groups, even if
the spatial distance was identical. These results did not
depend on whether judgments were based on perceptu-
ally available or memorized information, suggesting that
perceptual, not memory processes were responsible for
the formation of cognitive clusters.
Introduction
Spatial cognition is of central importance for a wide
range of human everyday activities, because most of our
actions rely on the perception and memories of spatial
relations, such as in reaching and grasping an object,
typing on a keyboard, or ®nding one's way home. To
perform successfully in such tasks, our cognitive system
not only needs to register and integrate relevant portions
of the available spatial information, but also retrieve and
use already acquired and stored information from short-
term and long-term memory. Interestingly, there is
strong evidence that spatial information undergoes
considerable changes on its way from the sensory surface
to memory, often distorting the original information in
systematic ways (for overview see McNamara, 1991;
Tversky, 1981). In the literature, such distortions have
been often attributed to memory processes, such as the
encoding of spatial information (e.g., McNamara &
LeSueur, 1989), its retrieval (e.g., Sadalla, Staplin, &
Burroughs, 1979), or both (Tversky, 1991). However,
here we entertain the hypothesis that at least some dis-
tortions might originate already from perception, not
memory, hence much earlier in the processing stream
that hitherto assumed. There is some evidence that
complex visual structures are coded in a hierarchical
fashion in both perception and memory. Memory dis-
tortions are often ascribed to hierarchical representa-
tion, so that such a commonality suggests that memory
distortions may merely re¯ect the perceptual organiza-
tion of stimulus information. We report two experiments
that investigated whether and how perceptual similari-
ties between perceived and to-be-memorized elements of
a map-like display aect perception- and memory-based
judgments of spatial relations. Our data provide evi-
dence that the structure of memory representations is
already formed in perception, a ®nding that calls for a
reinterpretation of a considerable part of previous
observations.
Hierarchical coding in memory
There are quite a number of studies supporting the idea
that spatial relations are coded hierarchically in memo-
ry. Maki (1981) had participants verify sentences de-
scribing the spatial relation between pairs of American
cities (``City A is west of City B'' or ``City A is east of
City B''), and observed that, as one might expect, veri-
®cation time was a decreasing function of Euclidean
inter-pair distance. However, this was only true for cities
that belonged to the same state (e.g., Alamo and
Psychological Research (2000) 64: 1±10 Ó Springer-Verlag 2000
B. Hommel (&)
University of Leiden, Section of Experimental and Theoretical
Psychology, Postbus 9555, 2300 RB Leiden, The Netherlands;
e-mail: hommel@fsw.leidenuniv.nl
B. Hommel á J. Gehrke á L. Knuf
Max Planck Institute for Psychological Research,
Munich, Germany
Burlington, North Dakota), but not for cities located in
dierent states (e.g., Jamestown, North Dakota, and
Albertville, Minnesota). Such ®ndings might indicate
that information about cities and states is hierarchically
organized, so that cities are stored as elements of su-
perordinate state categories. If so, comparing elements
from the same category should be in fact easier the more
discriminable (i.e., distant) the elements are; however,
judgments about elements from dierent categories
might be often based on category membership, hence
in¯uenced by the spatial relationship between categories
(i.e., states), so that within-category discriminability
does not (or not that much) come into play.
Further evidence for hierarchical structures in mem-
ory comes from the experiments of Stevens and Coupe
(1978). These authors presented their subjects with to-
be-memorized arti®cial maps each containing two cities
(e.g., city x and city y) that fell in dierent superordinate
regions (e.g., Alpha county, Beta county). In a congru-
ent condition, the spatial relation between the cities
matched the relation between the counties, e.g., city x
(located in Alpha county) was to the west of city y (lo-
cated in Beta county) and Alpha county was to the west
of Beta county. In an incongruent condition, the rela-
tionship between cities was the opposite of that between
counties, e.g., city x was to the west of city y and Alpha
county was to the east of Beta county. When subjects
made directional judgments about the two cities, sys-
tematic errors were observed with incongruent condi-
tions producing more errors than congruent conditions.
According to Stevens and Coupe, this is because par-
ticipants used their knowledge about superordinate re-
lations in judging the subordinated cities, so that the
judged relations were distorted to conform with the re-
lation of the superordinate geographical units.
A similar type of bias can also be demonstrated for
real-world locations, as shown by the study of Hirtle and
Jonides (1985) on the cognitive representation of land-
marks in the city Ann Arbor, Michigan, (e.g., city hall,
central cafe). Protocols of the free recall of landmarks
were used to (re-) construct individual clusters, sepa-
rately for each subject, and the validity of these clusters
was then tested by means of a spatial-judgment task
(i.e., distance estimation). As expected, distances within
a cluster were judged smaller than distances across
clusters.
In experiments reported by Hirtle and Mascolo
(1986), participants memorized maps in which place
names fell into two semantic cluster: names of recre-
ational facilities (e.g., Golf Course or Dock) and names
of city buildings (e.g., Post Oce or Bank). Locations
were arranged in such a way that, although places be-
longing to the same semantic cluster were spatially
grouped on the map, the Euclidean distance of one
recreational facility was shorter to the cluster of the city
buildings than to any other recreational facility, and vice
versa. However, when subjects were asked to estimate
inter-object distances on the basis of memory informa-
tion, they showed a clear tendency to (mis)locate these
critical places closer to their fellow category members
than to members of the other cluster.
Taken altogether, these results provide strong evi-
dence that global nonspatial relations between objects
induce the formation of hierarchical object clusters in
memory, thereby distorting certain inter-object spatial
relations, or at least the judgments made about these
relations.
Hierarchical coding in perception
The available results from memory studies provide
strong evidence for the assumption that information
about spatial con®gurations is not cognitively repre-
sented in a one-to-one correspondence, but seems to be
at least partly organized in a hierarchical fashion.
However, it is far from being settled which processes are
responsible for such an organization. Obvious candi-
dates are memory processes, which may work to reduce
the perceptual information to minimize storage costs,
optimize later retrieval, and so forth. But hierarchical
coding may also be a result of perceptual, or perceptu-
ally driven, processes, which may not only register sen-
sory evidence but actively integrate it into a structured
whole. If so, hierarchical coding in memory would tell us
not so much about memory principles but about per-
ceptual organization
1
.
Several authors have argued that complex visual
structures are perceptually coded in a hierarchical
fashion. Navon (1977) tested the idea that global
structuring of a visual scene precedes analysis of local
features. Participants were presented with large letters
(the global level) made of small letters (the local level),
and they were to recognize either the global or the local
letter level. There were two important outcomes: First, it
1
According to the classical Gestalt-psychological view, perceptual
organization proceeds preattentively and autonomously, hence
before codes of perceptually available information are brought into
contact with stored knowledge. If so, the rules according to which
perceptual information is organized are hard-wired into the per-
ceptual system and cannot be modi®ed by experience, an assump-
tion that draws a thick line between perceptual and memory
systems and processes. In contrast, more recent approaches (e.g.,
Goldstone & Barsalou, 1998) propose a more continuous view on
the relation between perception and memory that allows for in-
teractions between perceptual and memory contents in the pro-
cessing of perceptual input. We do not think that the logic of our
study presupposes the acceptance of one or other view, or that our
®ndings help to decide which one is more tenable. What we wish to
investigate is (1) whether the cognitive clustering (i.e., the organi-
zation) of spatial- and location-related information about map-like
layouts is aected, and can be experimentally induced, through
purely perceptual, nonspatial characteristics of the layout, and (2)
whether such clustering eects are restricted to memory-driven
tasks (i.e., tasks requiring memory retrieval) or whether they can
also be obtained before the layout has been memorized and stored,
i.e., in a perceptually-driven task not relying on memory retrieval.
However, even though our study does reveal clustering eects in
perceptually driven tasks ± an outcome suggesting an important
role of perceptual organization ± there is nothing in our data that
would exclude (or suggest) experience- or memory-related contri-
butions to perceptual organization.
2
took less time to identify the global than the local letter,
showing that global identi®cation is easier than local
identi®cation. Second, the congruence between global
and local letter produced asymmetric eects, that is,
global identi®cation was more or less independent of the
identity of the local letters, while local identi®cation was
much easier if global and local letters were identical than
if they were incongruent. This latter ®nding supports the
notion that local analysis is always preceded by global
processing, while global information can be extracted
without local analysis. Obviously, visual structures are
perceptually represented in a hierarchical fashion and
this hierarchy aects informational access.
More evidence for hierarchical clustering of visual
information has been found by Baylis and Driver (1993),
who had their subjects judge the relative height of object
features that were part of the same or of dierent visual
objects. Although the distance between the features was
held constant, the judgments were made faster when
both features were part of the same rather than dierent
objects. The authors argued that codes of features of the
same object, including their spatial relations, make up a
single representational cluster, with dierent clusters
(i.e., object representations) being hierarchically orga-
nized. If so, judging features of dierent objects requires
switching between cluster levels, while judging features
of the same object does not, so that between-level
judgments are slower than within-level judgments. Ob-
viously, these argument follow exactly the same lines as
those of Maki (1981), although Baylis and Driver refer
to perception, while Maki refers to memory. This
strengthens our suspicion that the way complex con®g-
urations are represented in perception and memory may
be similar, or even identical.
Gestalt principles in spatial cognition
Taken altogether, the literature reviewed so far suggests
that perceptual coding processes do not only aect
perceptually based judgments, but may also determine
the way perceptual information is stored, thus indirectly
aecting memory-based judgments. This implies that the
distortions and clustering eects observed in memory
tasks may not so much re¯ect organizational principles
of memory processes, but be a more or less direct con-
sequence of distortions and clustering tendencies in
perception. Accordingly, we wanted to test whether ex-
perimental, stimulus-related factors that are known to
aect perception can also be shown to aect memory-
based performance, and whether their eects on per-
ception and memory are comparable. In particular, we
investigated whether Gestalt principles can be demon-
strated to aect perception and memory the same way.
Gestalt principles are known to exert powerful eects
on perception. Important for our study is the fact that
elements tend to be grouped perceptually if they are
similar in color, shape, or other attributes (see Rock &
Palmer, 1990). Several investigators have shown that
such perceptual-clustering phenomena are associated
with biases and errors in spatial judgments analogous to
visual-geometric illusions (e.g., Canter & Tagg, 1975;
Coren & Girgus, 1980), suggesting that the perceived
distance between items belonging to the same phenom-
enal group is contracted. That is, Gestalt principles and
their eects on perceptual organization may lead to
distortions of the perceived space or, more precisely, of
the spatial characteristics of perceived con®gurations.
Indeed, Canter and Tagg found evidence that Gestalt
features of map-like stimuli can lead to measurable bi-
ases in performance: When facing a city with a river
running through it, people seemed to mentally divide the
city in halves, as expressed in their tendency to judge
distances between places within a half of the city dif-
ferently than across the river.
In the present two experiments, we used the Gestalt
principle of similarity, that is, the tendency of people to
perceive similar objects as belonging to a common
group. Similarity between objects was expected to in-
duce the cognitive clustering of the representations of
these objects, which again should aect performance in
perceptual and/or memory tasks. In Experiment 1, the
stimulus layout was a visual map-like con®guration of
18 objects (Fig. 1), namely houses of an arti®cial city.
The coloring of the objects was chosen in such a way
that the con®guration was subdivided into three or four
groups of adjacent objects with the same color. In Ex-
periment 2, the same grouping of objects on the stimulus
layout was achieved by using dierent shapes of the
objects (Fig. 2).
To test whether the non-spatial factors color and
shape would aect the coding of the spatial information
between the objects, participants were asked to perform
two ``spatial'' tasks: a speeded veri®cation of sentences
describing spatial relations (e.g., ``Is house A above
house B?'') ± a task often used in perceptual experiments
± and an unspeeded estimation of Euclidean distances ±
a task very common in memory experiments. We had
our participants perform these tasks under three condi-
tions in three consecutive sessions. In the perceptual
session, the con®guration was constantly visible; in the
memory condition, participants ®rst memorized the
con®guration and then performed the tasks without
seeing it; and in the perceptual/memory condition, the
con®guration was again visible, so that both perceptual
and memory information was available.
We expected both tasks to reveal the same pattern of
results, hence the cognitive clustering of the con®gura-
tion should aect veri®cation times as well as distance
estimations. Our prediction for both experiments were
as follows: Objectively identical Euclidean distances
between two given objects should be estimated as being
smaller when both objects were elements of the same
(identical color or shape) rather than dierent visual
groups. Such a result would suggest that the cognitive
representations of the objects were in fact clustered as a
consequence of Gestalt factors (e.g., similarity of color
or shape) and that this clustering led to the distortion of
3
the objective spatial information. We assumed that the
visual grouping of objects by similarity would induce a
similar kind of cognitive clustering as presenting or re-
ferring to abstract knowledge about category or group
membership ± the manipulation used in previous studies
(e.g., Stevens & Coupe, 1978). Along the same lines, we
expected the veri®cation of spatial relations to proceed
more quickly if the to-be-judged object pair belonged to
the same as compared to dierent visual groups. If so,
this would support the idea that (inter-)object informa-
tion is hierarchically represented, so that within-cluster
information can be accessed more quickly than between-
cluster information. Finally, given the idea that memory
eects may often merely re¯ect the outcome of percep-
tual organization, we expected that the same pattern of
result for distance estimations and veri®cation times
should be present in all three conditions, that is, results
to be independent of whether the perceptual information
is available or not.
Experiment 1
Experiment 1 was conducted to provide a ®rst test of our
hypothesis that Gestalt factors of a stimulus layout
(here: the color-based grouping of objects) induce the
hierarchical coding or clustering of the objects making
up this con®guration, and that this eect in¯uences
perception and memory in comparable ways.
Methods
Participants. Eighteen adults (mean age 24 years), 11 females and 7
males, were paid to participate in the experiment. They reported
having normal vision or corrected-to-normal vision, and were un-
aware of the purpose of the study.
Apparatus and stimuli. Stimuli were presented via a video beamer
(BARCODATA 800) on a 140 ´ 110 cm projection surface and
participants were seated in front of the surface with a viewing
distance of about 180 cm. The data acquisition was controlled by a
personal computer. Participants made their responses by pressing a
left or right sensor key with the corresponding index ®nger.
Stimuli were two versions of map-like con®gurations of 18
colored houses, which were introduced to the participants as houses
of a virtual city (Figs. 1, 2). The houses measured 15 ´ 15 cm and
were arranged as shown in Fig. 1, plus a randomly determined jitter
of up to 5 cm around the location centers. Each house was labeled
by a consonant-vowel-consonant nonsense syllable without any
obvious phonological, semantic, or functional relation to locations
or location-related words. The name-to-house mapping varied
randomly between participants. In one version of the con®guration
(C3), three dierent colors were used to subdivide the map into
three perceptual groups (group C3
1
: C, E, H, L, O; group C3
2
:B,F,
I, J, M, Q; and group C3
3
: D, G, K, N, P). In the second version
(C4) four groups were formed using four colors (group C4
1
:B,C,
D, F; group C4
2
: E, H, I, L; group C4
3
: G, J, K, N; and group C4
4
:
M, O, P, Q). The houses in locations A and R served as neutral
items in both con®gurations; their only use was to avoid possible
end or anchor eects on veri®cation or estimation performance.
Eight vertical location pairs were chosen for location veri®ca-
tions and distance estimations. The members of each of these pairs
were separated by 300 mm on average. Half the pairs were com-
posed of elements which were located within a perceptual group
(B±F, E±L, G±N, M±Q) and the other half consisted of elements of
adjacent groups (C±I, D±J, I±O, J±P). A small set of diagonal and
Fig. 1 Schematic graph of the layout used in Experiment 1
(example). The letters indicating the locations were not shown to
the subjects; instead each house was identi®ed by a nonsense
``name'' (i.e., a meaningless syllable like ``MAW'', omitted here)
appearing in its center. Each con®guration consisted of 18 colored
houses (colors indicated here by texture). Three to four of them had
the same color, this making up either three or four color groups
(con®gurations C3 and C4). For example, in C4 houses in locations
B, C, D and F might be red, houses in E, H, I, and L green, houses
in G, J, K, and N blue, and houses in M, O, P and Q yellow. White
houses A and R, that were not used in any task, were added to all
con®guration versions
Fig. 2 Illustration of the objects used in Experiment 2. All objects
were uncolored and varied in shape only. Three to four of them
were of the same type, this making up either three or four shape
groups (con®gurations C3 and C4, respectively). Houses of type 3
served as neutral objects in locations A and R
4
horizontal pairs was used as ®llers, but results for these pairs were
not further analyzed.
Design. The experiment consisted of three experimental sessions
(perceptual, memory, and perceptual/memory condition). Each
session was divided into one experimental block for location
judgments and another block for distance estimations, with task
order being balanced across participants. A set of 320 judgments
was composed of 10 repetitions for each possible combinations of
eight experimental pairs, two relations (under, above), and two
orders of location within the pair (A±B, B±A); 88 ®ller judgments
were added to the set. Half of the participants responded yes and
no by pressing the left and right response key, respectively, while
the other half received the opposite response-key mapping. A set of
48 distance estimations comprised three repetitions of each of the
possible combinations of eight experimental pairs and two order-
ings of location within the pair. Twelve further pairs served as
®llers. Half of the participants worked on con®guration C3, con-
sisting three perceptual groups, and the other half worked on
con®guration C4, consisting four perceptual groups.
Procedure. Each participant participated in three experimental
sessions on three consecutive days. The stimulus con®guration for a
given participant was the same in each session. In the ®rst session
(perceptual condition), the con®guration remained visible while
subjects worked through the distance estimation task and the lo-
cation veri®cation task. The second session (memory condition)
started with an acquisition phase, in which the participants mem-
orized the positions and syllables of the displayed houses. Then
they performed the estimation and the veri®cation tasks in front of
a blank projection surface, hence without seeing the con®guration.
The third session (perceptual/memory condition) was identical to
the ®rst session, hence the con®guration was visible while the two
tasks were performed.
Distance estimations. Sixty pairs of house names (48 critical dis-
tance pairs and 12 ®ller pairs) were displayed one pair at a time in
the upper center of the projection surface. The names were dis-
played in adjacent positions, separated by a short horizontal line.
Another horizontal line of 70 cm in length was shown below the
names and participants were told that this line represented the
width of the whole projection surface (which actually spanned
double the size). It was crossed by a vertical pointer of 5 cm in
length, which could be moved to the left or right by pressing the left
and right response key, respectively. For each indicated pair, par-
ticipants were required to estimate the distance between the cor-
responding objects (center to center) by adjusting the location of
the pointer accordingly, and then to verify their estimation by
pressing the two response keys at the same time. They were in-
structed to take as much time as needed for each estimation. The
critical dependent measure (i.e., estimated distance) was computed
by taking the distance indicated on the screen in pixels, multiplied
by 2, and transforming it into millimeters.
Location judgments. A series of 408 (320 critical and 88 ®ller) to-
be-veri®ed locational statements was presented to each participant,
one statement at a time. In each trial, a ®xation cross appeared for
300 ms in the top center of the display. Then the statement ap-
peared, consisting of the names of two objects and a relation be-
tween them, such as ``RUK under JOX'' or ``KAD above NOZ''.
Participants were instructed to verify (or falsify) as quickly and as
accurately as possible by pressing the yes or no key accordingly; the
assignment of answer and response key was counterbalanced across
participants. The sentence stayed on the projection surface until the
response key was made. After an intertrial interval of 600 ms the
next trial appeared. In the case of an incorrect key being pressed,
the error was counted without feedback and the trial was indexed.
Such an indexed trial was incorporated into the rest of the series
at a random position within the series. If the same error on the
same trial was made for three times, this trial was excluded from
the data.
Acquisition. The second session always started with the acquisition
of the stimulus con®guration. The con®guration was presented to
the participants, who had unlimited time to memorize the locations
and names of the displayed objects. The con®guration then dis-
appeared and the participants were sequentially tested for each
object. An object-size rectangle appeared in the lower right corner
of the display, together with an object name in the lower left corner.
Using a joystick, participants were asked to move the rectangle to
the exact position of the named object. After pressing the left and
right key simultaneously, the computer recorded the position of the
rectangle, the projection surface was cleared, and the next test trial
started. There were 18 such trials, one for each object, in a random
order. If an object was mislocated for more than 2.5 cm, the whole
procedure was repeated from the start
2
.
The acquisition phase ended after the participant completed
three correct positioning procedures.
Results
From the data of the distance-estimation task, mean
estimates in millimeters were computed for each partic-
ipant and condition. Over all conditions, the real dis-
tance was clearly underestimated: 237 mm instead of
300 mm. Estimates took about 15 s on average and
there was no indication of any dependency of estimation
latency on session. To test our hypotheses, comparisons
were made between pooled estimates of the within-group
pairs B±F, E±L, G±N and M±Q and pooled estimates of
the between-groups pairs C±I, D±J, I±O and J±P.
However, an ANOVA with the within-subject factors
session/condition (perceptual, memory and perceptual/
memory) and group membership (within-group vs be-
tween-groups) did not reveal any signi®cant main eect
or interaction. That is, no systematic distortions were
observed for object pairs spanning two vs one group.
Figure 3 shows the estimated distances for the pooled
within-group pairs and between-groups pairs across
sessions.
In the locational-judgment task, error rates were be-
low 5% and the respective trials were excluded from
2
This procedure might have provided subjects with feedback to
improve and correct their spatial memory, which again might have
worked against possible distortion eects. However, there are
reasons why this should not represent a serious problem for in-
terpreting our ®ndings. First, every study on spatial memory re-
quires some kind of check of whether the stimulus con®guration
has actually been learned. The previous studies described above
also used some kind memory test that was repeated if too many
errors were made. Despite this feedback, several of these studies
found distortions in distance estimations. Second, even if repeti-
tions of the test were used as feedback, the subjects could not
attribute their failure to a speci®c object. Therefore, all that the
repetitions oered was practice, which should have aected all
objects ± and, by implication, all object pairs ± equally. This rules
out speci®c eects of test repetitions on represented locations or
distances between within- and/or between-groups pairs ± apart
from the fact that a tolerance of 25 mm still leaves room for some
distortion eects.
5
analysis. In a ®rst step, the remaining reaction times
were analyzed as a function of con®guration type (C3 vs
C4), spatial relation (above vs under), and response (yes
vs no). The corresponding ANOVA revealed a highly
signi®cant main eect for response, F(1, 16) 39.98,
p < 0.001, indicating that positive responses were faster
(2,656 ms) than negative responses (2,888 ms). The sec-
ond signi®cant source of variance was an interaction of
relation and response, F(1, 16) 5.24, p < 0.05. As
revealed by a Schee
Â
test (p < 0.05), this was produced
by an eect of relation restricted to positive reactions,
where subjects veri®ed ``above'' statements faster than
``below'' statements (2,588 vs 2,724 ms). Other sources
of variance failed to reach the signi®cance level.
Again, the most interesting analysis concerned the
comparison of the reaction times within- and between-
groups pairs (Fig. 4). In an ANOVA with the factors
session/condition and group membership, the main ef-
fects of session/condition, F(2, 16) 38.74, p < 0.01,
and of group membership, F(1, 17) 54.36, p < 0.01,
were signi®cant, while the interaction was not. In
contrast to the distance estimation task, the grouping of
the elements by color was quite eective: Faster reac-
tion times were obtained within groups of identical
colored objects (2,523 ms) than between groups
(3,021 ms). Additionally, the reaction times decreased
over sessions (critical dierence of the Schee
Â
test:
661 ms, p < 0.05), but the type of session did not aect
the grouping eect.
Discussion
The aim of Experiment 1 was to examine how a visual
layout of objects is coded in perception and memory
and, in particular, whether the color-induced grouping
of objects leads to the hierarchical organization of their
cognitive representations. We used two tasks which are
very common in perception and memory studies. In
contrast to our expectations of converging results in
both tasks, the results turned out to depend on the
measure taken.
Fig. 3 Mean estimated Euclid-
ean inter-object distance in
Experiment 1 as a function of
sessions/conditions (P percep-
tual condition, M memory
condition, P/M perceptual/
memory condition) and group
membership of object pairs.
The dotted line indicates real
distances
Fig. 4 Mean reaction times of
judged spatial propositions in
Experiment 1 as a function of
session/condition (P perceptual
condition, M memory condi-
tion, P/M perceptual/memory
condition) and group member-
ship of object pairs
6
On the one hand, performance in the veri®cation task
provided some strong evidence for hierarchical organi-
zation: Reaction times were shorter if the to-be-judged
object pair belonged to the same rather than to dierent
color groups. This suggests that objects of the same
color were integrated into one cognitive cluster, and that
this similarity-based clustering aected the access to
representations of cluster members in judgments of rel-
ative location. Importantly, pronounced within-group
bene®ts were found in all three sessions, that is, inde-
pendent of whether perceptual information was avail-
able or not. This observation has two implications.
First, the fact that clustering eects appear in the
®rst, perceptual session already demonstrates that these
eects do not require memory storage and retrieval to
show up. Apparently, the cognitive structuring of in-
formation can be induced by purely perceptual means
like visual similarity manipulations. As mentioned
above, this does not necessarily mean that the clustering
is achieved by some autonomous, purely perceptual
system without any top-down in¯uence from memory ±
nor does it exclude this possibility. However, it does
imply that when people perceive a relatively complex
visual layout, they already integrate and cluster the
perceptually available information in ways that show up
in their location judgments. Second, the fact that the size
of clustering eects does not increase if the judgments
are made from memory at least suggests that processes
of memory encoding and/or retrieval do not determine,
and perhaps do not even contribute to, that and how the
spatial information is coded and integrated. However,
this second conclusion needs to be treated with caution
as it rests on a statistical null eect (the absence of a
session-by-group interaction). Moreover, Experiment 2
will show that there are situations where the size of the
grouping eect can vary between sessions. What seems
clear, though, is that, if anything, storage and retrieval
adds very little to the clustering eect, suggesting that
perceptual and memory-informed judgments are based
on the same cognitive representations.
On the other hand, visual grouping did not produce
any systematic distortion in distance estimations. Apart
from the general tendency to underestimate physical
distances, no dierences could be observed between the
estimates of within-group and between-groups pairs.
There are at least four possible explanations for this
®nding. First, the absence of grouping eects on distance
estimations might indicate that grouped objects were
simply not cognitively clustered. Given that eects in-
dicative of clustering were obtained in the veri®cation
latencies, this explanation is not very plausible unless
one could come up with a reasonable story of why the
two tasks were aected dierently.
Second, and this may be such a story, estimating a
distance and verifying a location require dierent types
of judgments and make use of dierent measurement
scales. Therefore, a dissociation of these two measures
may simply re¯ect their dierent degrees of sensitivity to
memory distortions. Although this possibility cannot be
ruled out on the basis of the present ®ndings, it would
raise the question of why a discrete, relative judgment
should be more sensitive than a continuous, absolute
judgment that has already been shown to be aected by
memory distortions in several previous studies.
Third, objects and their relations might have been
represented in a distorted way but, in contrast to the
veri®cation task, the unspeeded nature of distance esti-
mations allowed for correcting these errors before overt
emission of the response
3
.
Fourth, it may also be that veri®cation latency and
the reliability of distance estimations tap dierent cog-
nitive processes. Distance estimations may more or less
directly re¯ect the quality and representation of spatial
information that is used in both veri®cation and esti-
mation tasks. If so, the fact that we did not ®nd sys-
tematic distortions of estimations would indicate that
this information was relatively reliable (i.e., of high
quality) at least in our study. Yet, veri®cation latencies
need not re¯ect the quality of spatial information. In-
stead, they may depend on how accessible this infor-
mation is, which again depends on how it is stored and
organized. That is, Gestalt-induced grouping may lead
to the hierarchical clustering of data entries (i.e., the
formation of stronger links between codes of objects
belonging to the same visual group) but not, or not
necessarily, to the distortion of the data themselves.
On the basis of our results we are not able to dis-
tinguish between the last three of these possibilities.
Importantly, however, each suggests that distance esti-
mations and veri®cations of spatial relations measure
dierent processes (or at least measure them to dierent
degrees) and, therefore, tap into dierent aspects of
spatial representations. Given that in the past distance-
estimation and location-veri®cation tasks have been
used more or less interchangeably, such a possibility
raises important implications for further research on
spatial cognition apart from and beyond the present
study.
The dissociation of veri®cation and estimation per-
formance notwithstanding, it is interesting to note that
there were no systematic dierences between estimations
under perceptual and memory conditions, hence no
main eect of session. In contrast to this ®nding, pre-
vious research has suggested that judgments made from
the memory can dier systematically in the accuracy
from those made perceptually (Kerst & Howard, 1978;
Moyer, Bradley, Sorensen, Whiting, & Mans®eld, 1978).
For example, Kerst, Howard, and Gugerty (1987) re-
ported that perceptual pair-distance judgments were
more veridical than judgments made from memory after
a brief (10 min) or a long (24 h) retention period.
3
An obvious test of this idea would be to vary the time available for
estimations. We actually did this in a follow-up experiment, where
estimation time was either limited or unlimited ± with little success
(i.e., no impact on the estimations). However, as even in the limited
condition the interval was as long as 7 s (which was the minimum
to handle the input device), we do not consider this to represent a
strong test of the correction hypothesis.
7
Interestingly, however, the properties of the used maps
itself ± the building sites on the left half of the map were
labeled with dormitory names and those on the right
were labeled with department names ± had no eect on
the accuracy of distance judgments. This latter obser-
vation is in line with our ®ndings, thus supporting the
view of a close correspondence between perception and
memory-based estimates and, presumably, of the data
these processes operate on.
Experiment 2
Experiment 1 provides the ®rst evidence that nonspatial
Gestalt factors like color grouping aect the coding and/
or organization of spatial information in the perception
and memory of map-like layouts. Also of interest, al-
though the two measures we used yielded dierent out-
comes, there was no evidence for session or condition
eects, suggesting that perceptual and memory processes
are based on the same representations. Experiment 2 was
conducted to replicate and extend Experiment 1 for
three reasons. First, we wanted to see whether our basic
®nding generalizes to other domains, i.e., whether a
within-group bene®t in veri®cation times can be ob-
served with another grouping factor than color. Second,
given that the somewhat surprising dissociation of ver-
i®cation and estimation measures was based on a null
eect in the estimation task, we were interested in
whether this null eect can be replicated at all. Third, it
seemed important to test whether the null interaction of
group membership and session/condition can be repli-
cated. After all, arguing on the basis of a null eect is
problematic, so that we sought for converging evidence
to strengthen our conclusions from Experiment 1. As a
result of these considerations, we repeated Experiment 1
but, instead of color, attempted to induce similarity
between objects by shape. That is, visual groups were
formed using the same type of house for each subgroup
of objects.
Method
Eighteen dierent adults (mean age 24 years), 11 females and 7
males, were paid to participate in the experiment. They ful®lled the
same criteria as in Experiment 1. The apparatus was the same as in
Experiment 1 as was the method, except that the 18 houses diered
not in color (i.e., all were presented in black outline on white
background) but shape (see Fig. 2).
Results
The data were analyzed as in Experiment 1. In the dis-
tance-estimation task, the real distance was again un-
derestimated: 222 mm instead of 300 mm. Estimates
took about 12 s on average, with no indication of any
dependency of estimation latency on session. An
ANOVA revealed only a signi®cant main eect for the
factor session/condition, F(2, 15) 5.96, p < 0.01, due
to participants estimating the Euclidean distances be-
tween objects as being shorter in the memory condition
(217 mm) than in the perceptual condition (240 mm;
critical dierence of Schee
Â
test: 17 mm, p < 0.05)
(Fig. 5).
In the locational-judgment task, error rates were
again below 5%. An ANOVA with the factors con®g-
uration type (C3 vs C4), spatial relation (above vs un-
der), and response (yes vs no) revealed main eects of
spatial relation, F(1, 16) 13.57, p < 0.01, and re-
sponse, F(1, 16) 42.28, p < 0.01. Mean reaction times
were shorter if a yes response was required (3,257 ms) as
compared to a no response (3,550 ms). Furthermore,
``above'' judgments were faster (3,341 ms) than ``below''
judgments (3,466 ms). Other sources of variance failed
to reach signi®cance.
A further ANOVA with the factors session/condition
and group membership yielded a highly signi®cant main
eect of session/condition, F(2, 16) 84.97, p < 0.01,
which showed a decreasing reaction time over sessions
(perceptual condition: 4,272 ms; memory condition:
3,571 ms; perceptual/memory condition: 2,447 ms; crit-
ical dierence of the Schee
Â
test: 489 ms, p 0.05). The
Fig. 5 Mean estimated Euclid-
ean inter-object distance in
Experiment 2 as a function of
sessions/conditions (P percep-
tual condition, M memory
condition, P/M perceptual/
memory condition) and group
membership of object pairs.
The dotted line indicates real
distances
8
second signi®cant source of variance was the main group
membership eect, F(1, 17) 15.34, p < 0.01, that
replicated the results from Experiment 1: faster reaction
times for within-group pairs (3,165 ms) than for be-
tween-groups pairs (3,695 ms). This time, however, the
two factors interacted, F(2, 16) 4.69, p < 0.05, due to
larger group-membership eects in the second and third
session than in the ®rst session. However, separate
comparisons con®rmed that the group-membership ef-
fect was signi®cant in all sessions, p < 0.05 (Fig. 6).
Discussion
All in all, Experiment 2 represents a successful replica-
tion of Experiment 1. With regards to our ®rst goal, this
means that eects of inter-object similarity on veri®ca-
tion performance is not limited to the domain of color
but can also be demonstrated for shape. Although other
types of similarity may also be interesting to investigate,
we can safely conclude that the grouping eect seems to
be fairly general. A second aspect of the present results is
that the selective eect of group membership on veri®-
cation times, but not on distance estimations, could be
replicated as well. Even though one may still argue that
the null eect in the estimation task merely re¯ects its
less pronounced sensitivity, the replication strengthens
our suspicion that veri®cation and estimation tasks
measure dierent things.
With regard to our third goal, the outcome is
somewhat mixed. On the one hand, grouping aected
performance from the ®rst session on, which points again
to a major role of cognitive clustering in a perceptually
driven task. On the other hand, in contrast to Experi-
ment 1, the eect of grouping on veri®cation latencies
increased from the ®rst, perceptual session to the two
following, memory-related sessions. There are at least
three possible, partly related reasons for this increase.
First, it may be that storing and/or retrieving infor-
mation about spatial layouts as such introduces further
integration processes that strengthen the organization
suggested or already achieved by perceptual or percep-
tually driven processes. Second, the need to memorize a
complex spatial layout may lead subjects to make active,
strategic use of the structured organization already of-
fered by their perceptual system. Third, performance in
the ®rst, perceptual session may re¯ect a mixture of
several, sometimes opposing eects of perceptual orga-
nization. Although our shape manipulation was quite
eective, we cannot exclude the additional impact of
other organizational tendencies. For instance, people
may tend to organize our layouts into diagonal clusters,
this way grouping, say, locations A, B, D, G, and K,
locations C, F, J, and N, and so forth. Accordingly,
the pair B±F, say, would represent a between-groups
pair, which would con¯ict with the shape-induced cod-
ing as a within-group pair. Once coded and stored,
however, one interpretation ± presumably the shape-
based one ± might prevail, so that no coding con¯ict
would occur from session 2 on. From this view, the in-
teraction of session and group membership may merely
re¯ect the fact that perceptual information can be or-
ganized in dierent ways.
At this point, we are unable to decide between these
three possibilities. On the one hand, the fact that clus-
tering eects varied with session in Experiment 2 but
not in Experiment 1 seems to point to a more strategy-
based interpretation, such as the second, memorizing-
related hypothesis. On the other hand, it seems clear
that a ®rmer conclusion regarding this issue would
presuppose a direct manipulation of the individual en-
coding and/or retrieval strategies. Nevertheless, the
most important message from Experiment 2 is that,
whatever may have increased the clustering eect in the
second session, it was reliably present before. This
demonstrates that the hierarchical organization of in-
formation about map-like layouts appears in perceptu-
ally driven performance already and, therefore, does not
require or rely on processes responsible for memory
storage or retrieval.
Fig. 6 Mean reaction times of
judged spatial propositions in
Experiment 2 as a function of
session/condition (P perceptual
condition, M memory condi-
tion, P/M perceptual/memory
condition) and group member-
ship of object pairs
9
Conclusions
Studies of human spatial cognition commonly focus on
either the on-line use of perceptual information for on-
going action or on the representation and organization
of spatial knowledge in memory. In contrast, the present
experiments investigated and compared the ways spatial
information is coded in perception and memory. In our
view, three important conclusions can be drawn from
our ®ndings.
First, exogenous factors such as Gestalt characteris-
tics of spatial layouts strongly aect the way spatial in-
formation is organized. In showing that, our results are
in line with, and extend, previous demonstrations that
spatial memories are in¯uenced by nonspatial informa-
tion such as linguistic information (Bower, Karlin, &
Dueck, 1975; Daniel, 1972), semantic relations (Hirtle &
Mascolo, 1986; McNamara & LeSueur, 1989; Sadalla
et al., 1979), functional object information (McNamara,
Halpin, & Hardy, 1992), or landmark-induced visual
grouping (Gehrke & Hommel, 1998). These demonstra-
tions are inconsistent with the idea that cognitive repre-
sentations are mere copies of external stimulus arrays.
Instead, spatial information seems to be integrated with
nonspatial information in such a way that the retrieval of
spatial information leads to the automatic activation of
nonspatial information and, presumably, vice versa.
Second, there was very little evidence, if any, for
coding dierences in perception and memory. In Ex-
periment 1 color grouping aected perception- and
memory-based performance to the same degree, and the
dierences observed in Experiment 2 were more quan-
titative than qualitative in kind. In as much as exoge-
nous grouping can be assumed to induce the cognitive
clustering of spatial information, this implies that clus-
tering arises in perception already instead of re¯ecting a
process of memory organization. Accordingly, at least
part of the demonstrated in¯uences of nonspatial factors
on spatial memory may tell us more about the principles
of coding and organization in the perceptual system
than about memory processes.
Third, the lack of converging results in the location-
veri®cation and the distance-estimation tasks raises the
question of whether both tasks tap into dierent pro-
cesses. Although we are reluctant to draw strong con-
clusions from the present data, it seems clear that these
two measures should not be treated as equivalent. Pos-
sibly, distance estimations assess the quality of spatial
information, while veri®cation latencies indicate the way
this information is organized.
Acknowledgements The research reported in this paper was sup-
ported by a grant of the German Science Foundation (DFG, HO
1430/6-1/2) and by the Max Planck Institute for Psychological
Research in Munich. We are grateful to Sabine Miller, Edith
Mu
È
ller, and Agnes Pohlinger for collecting the data.
References
Baylis, G. C., & Driver, J. (1993). Visual attention and objects:
evidence for hierarchical coding of location. Journal of Exper-
imental Psychology: Human Perception and Performance, 19,
451±470.
Bower, G. H., Karlin, M. B., & Dueck, A. (1975). Comprehension
and memory for pictures. Memory and Cognition, 3, 216±220.
Canter, D., & Tagg, S. K. (1975). Distance estimation in cities.
Environment and Behavior, 7, 59±80.
Coren, S., & Girgus, J. S. (1980). Principles of perceptual organi-
zation and spatial distortion: the gestalt illusions. Journal of
Experimental Psychology: Human Perception and Performance,
6, 404±412.
Daniel, T. C. (1972). Nature of the eect of verbal labels on rec-
ognition memory for form. Journal of Experimental Psychology,
96, 152±157.
Gehrke, J., & Hommel, B. (1998). The impact of exogenous factors
on spatial coding in perception and memory. In: C. Freksa,
C. Habel, & K. F. Wender (Eds.), Spatial cognition: an inter-
disciplinary approach to representing and processing spatial
knowledge (pp. 63±78). Berlin, Germany: Springer.
Goldstone, R. L., & Barsalou, L. (1998). Reuniting perception and
conception. Cognition, 65, 231±262.
Hirtle, S. C., & Jonides, J. (1985). Evidence of hierarchies in cog-
nitive maps. Memory and Cognition, 13, 208±217.
Hirtle, S. C., & Mascolo, M. F. (1986). Eect of semantic clustering
on the memory of spatial locations. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 12, 182±189.
Kerst, S. M., Howard, J. H. (1978). Memory psychophysics for
visual area and length. Memory and Cognition, 6, 327±335.
Kerst, S. M., Howard, J. H., Gugerty, L. J. (1987). Judgment ac-
curacy in pair-distance estimation and map sketching. Bulletin
of the Psychonomic Society, 25, 185±188.
Maki, H. (1981). Categorization and distance eects with spatial
linear orders. Journal of Experimental Psychology: Human
Learning and Memory, 7, 15±32.
McNamara, T. P. (1991). Memory's view of space. Psychology of
Learning and Motivation, 27, 147±186.
McNamara, T. P., Halpin, J. A., Hardy, J. K. (1992). The repre-
sentation and integration in memory of spatial and nonspatial
information. Memory and Cognition, 20, 519±532.
McNamara, T. P., LeSueur L. L. (1989). Mental representations of
spatial and nonspatial relations. Quarterly Journal of Experi-
mental Psychology, 41, 215±233.
Moyer, R. S., Bradley, D. R., Sorensen, M. H., Whiting, J. C., &
Mans®eld, D. P. (1978). Psychophysical functions for perceived
and remembered size. Science, 200, 330±332.
Navon, D. (1977). Forest before trees: the precedence of global
features in visual perception. Cognitive Psychology, 9, 353±383.
Rock, I., & Palmer, S. (1990). The legacy of Gestalt psychology.
Scienti®c American, 263, 48±61.
Sadalla, E. K., Staplin, L. J., & Burroughs, W. J. (1979). Retrieval
processes in distance cognition. Memory and Cognition, 7, 291±
296.
Stevens, A., & Coupe, P. (1978). Distortions in judged spatial re-
lations. Cognitive Psychology, 10, 422±427.
Tversky, B. (1991). Spatial mental models. Psychology of Learning
and Motivation, 27, 109±145.
10