Psychological Science

23(2) 120 –125

© The Author(s) 2012

Reprints and permission:

sagepub.com/journalsPermissions.nav

DOI: 10.1177/0956797611429467

http://pss.sagepub.com

Unlike navigating unfamiliar terrain, navigating one’s city of

residence is usually an error-free and effortless endeavor. In

the research reported here, we investigated how such highly

familiar spaces are represented in memory. Several research-

ers have proposed that memory for highly familiar spaces is

independent of orientation (Byrne, Becker, & Burgess, 2007;

Evans & Pezdek, 1980; Gallistel, 1990; Sholl, 1987) and that

navigation based on such orientation-free memory does not

depend on navigators’ orientation within these environments.

In contrast, some theories assume that spatial memory is ori-

entation dependent and that locations in a familiar city are rep-

resented within one oriented global reference frame (GRF;

McNamara, Sluzenski, & Rump, 2008; O’Keefe, 1991;

Poucet, 1993; Trullier, Wiener, Berthoz, & Meyer, 1997).

Therefore, performance in survey tasks (tasks that focus on the

metric relations between mutually nonvisible locations; e.g.,

pointing, distance-estimation, or shortcutting tasks) might rely

on this oriented GRF.

A global alignment effect, whereby performance is best

when body orientation (i.e., viewing direction) is parallel to

the orientation of the GRF representing the navigated space

(Iachini & Logie, 2003; Levine, Marchon, & Hanley, 1982;

McNamara et al., 2008), is evidence for the use of an oriented

GRF in survey tasks. When body orientation and GRF

orientation are not parallel, strategies for realignment (e.g.,

mental rotation) may lead to reduced performance. Although

performance generally decreases with increasing misalign-

ment (Iachini & Logie, 2003), orthogonal and contra-aligned

(i.e., rotated by 180°) body orientations often yield better per-

formance than do oblique misalignments (McNamara et al.,

2008).

If participants rely on an oriented navigation-based GRF in

survey tasks, the alignment effect should be identical for all

target locations represented within one GRF. The orientation

of this GRF should be determined by both environmental

structure and navigators’ individual experiences; because of

their unique individual experiences, different navigators are

likely to conceive differently oriented GRFs. If all target loca-

tions are represented within one navigation-based GRF, a par-

ticipant should be able to access all target locations with equal

Corresponding Authors:

Julia Frankenstein, Center for Cognitive Science, University of Freiburg,

Friedrichstrasse 50, 79098 Freiburg, Germany

E-mail: julia.frankenstein@cognition.uni-freiburg.de

Tobias Meilinger, Max Planck Institute for Biological Cybernetics,

Spemannstrasse 38, 72076 Tübingen, Germany

E-mail: tobias.meilinger@tuebingen.mpg.de

Is the Map in Our Head Oriented North?

Julia Frankenstein1,2, Betty J. Mohler1, Heinrich H. Bülthoff1,3, and

Tobias Meilinger1

1Max Planck Institute for Biological Cybernetics, Tübingen, Germany; 2Center for Cognitive Science, University of

Freiburg; and 3Department of Brain and Cognitive Engineering, Korea University

Abstract

We examined how a highly familiar environmental space—one’s city of residence—is represented in memory. Twenty-six

participants faced a photo-realistic virtual model of their hometown and completed a task in which they pointed to familiar

target locations from various orientations. Each participant’s performance was most accurate when he or she was facing

north, and errors increased as participants’ deviation from a north-facing orientation increased. Pointing errors and latencies

were not related to the distance between participants’ initial locations and the target locations. Our results are inconsistent

with accounts of orientation-free memory and with theories assuming that the storage of spatial knowledge depends on local

reference frames. Although participants recognized familiar local views in their initial locations, their strategy for pointing relied

on a single, north-oriented reference frame that was likely acquired from maps rather than experience from daily exploration.

Even though participants had spent significantly more time navigating the city than looking at maps, their pointing behavior

seemed to rely on a north-oriented mental map.

Keywords

spatial memory, environmental space, reference frame, local and global reference frames, orientation, alignment, map,

virtual reality

Received 2/1/11; Revision accepted 8/5/11

Research Report

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from

The Orientation of Mental Maps 121

speed and precision. However, oriented GRFs constructed

from exploration may contain errors and distortions that accu-

mulate as the navigation distance, and thus the size of the rep-

resented area, increases (Loomis et al., 1993). Increasing the

distance between pointing locations and targets may thus yield

greater pointing errors (an effect known as a distance effect).

Other theoretical frameworks (Christou & Bülthoff, 1999;

Gillner, Weiß, & Mallot, 2008; Meilinger, 2008; Wang &

Spelke, 2002) propose that spatial knowledge is stored using

local reference frames (LRFs). LRFs correspond to sur-

roundings that are usually visible from a single vantage

point, such as a street or a town square. The orientation of

LRFs is derived from local geometry (e.g., street orienta-

tion), navigators’ experienced perspective, or both. When

individuals perform survey tasks, they integrate multiple

LRFs into a single reference frame; Meilinger (2008) pro-

posed that this integration is based on the LRF of the naviga-

tor’s current (real or imagined) position. According to this

account, participants should perform survey tasks best when

their body position is aligned with the orientation of a local

street, because people encode LRFs parallel to streets while

walking. In addition, because pointing to more distant loca-

tions requires the integration of more LRFs, reliance on

LRFs in a pointing task should result in longer latencies and

larger errors (i.e., distance effects).

Spatial relations can be learned not only from experience

but also from maps (Richardson, Montello, & Hegarty, 1999;

Sun, Chan, & Campos, 2004; Thorndyke & Hayes-Roth,

1982). If spatial relations are learned from maps, then Western

participants, who typically use maps that display locations

within a north-oriented GRF, should perform best when they

are facing north, and their performance should decrease as

their angle of misalignment from this north-facing orientation

increases. If participants rely on a map-based GRF in a point-

ing task, no distance effects should arise, because map-based

memory for nearby and distant locations should not differ in

accessibility or precision.

By observing the effects of alignment relative to LRFs

(parallel to streets) and oriented GRFs (north-oriented and

individual) on performance in a survey task, we sought to

assess the spatial encoding strategy of individuals. If individu-

als use orientation-free representations, then no alignment

effect should occur in such a task. If individuals use LRFs,

both pointing errors and pointing latencies should show dis-

tance effects. However, if individuals use oriented navigation-

based GRFs, distance effects should emerge for pointing errors

but not for pointing latencies. If individuals rely on oriented

map-based GRFs, no distance effects should occur.

To test these predictions, we conducted an experiment

using a novel pointing task. Participants wearing head-

mounted displays faced five familiar locations (initial loca-

tions) in a virtual model of their hometown (see Fig. 1) and

were asked to point to target locations that were not visible to

them. We examined performance as a function of body orien-

tation and target distance.

Method

Participants

Twenty-seven naive participants (14 male, 13 female), ages 18

to 50 years (M = 28.5 years, SD = 7.7), were recruited from a

subject database and took part in our experiment in exchange

for monetary compensation. All participants had lived in

Tübingen, Germany, for at least 2 years (M = 6.7 years, SD =

5.4). (One additional participant did not complete the experi-

ment and was excluded from analysis.)

Apparatus and materials

We used a highly realistic virtual model of Tübingen (see

Fig. 1; Meilinger, Knauff, & Bülthoff, 2008; the virtual model

can be viewed online at http://virtual.tuebingen.mpg.de). Par-

ticipants saw the model in ground perspective through a head-

mounted display while sitting on a tall chair. Simulated fog in

the virtual model ensured similar viewing depths in all direc-

tions. The experiment was programmed using Virtools (Ver-

sion 4.0; Dassault Systemes, Vélizy-Villacoublay, France).

To render an egocentric view of the virtual environment in

the head-mounted display in real time, we tracked participants’

head coordinates using four high-speed Vicon MX 13 motion-

capture cameras (Vicon, Los Angeles, CA) with a frame rate

of 120 Hz. Displays were generated with a Nvidia GO 6800

Ultra graphics card (Nvidia, Santa Clara, CA) with 256 MB of

RAM; we used a Kaiser SR80 head-mounted display (Kaiser

Electro-Optics, Carlsbad, CA) with a field of view of 63° (hor-

izontal) × 53° (vertical) and a resolution of 1,280 × 1,024 pix-

els for each eye. The interpupillary distance was fixed at 8 cm.

We adjusted the fit of the head-mounted display and the place-

ment of the display screen individually for each participant.

The setup of the apparatus provided important depth cues,

such as stereo vision and motion parallax. We measured par-

ticipants’ pointing performance using a custom-made joystick

with a resolution of approximately 2°.

Procedure

Locations in the model consisted of a castle courtyard, three tav-

erns, a train station, a fire station, a mall, a museum, a cinema,

three intersections, and a university building. On each trial, par-

ticipants faced an initial location from a specific orientation. Par-

ticipants first self-localized (i.e., they confirmed that they

recognized the location and orientation) by pressing a button.

They then pointed in the virtual direction of three specific target

locations whose written names appeared separately on the screen

of the head-mounted display. Participants were free to rotate

while they self-localized, but while they pointed, we enforced a

fixed head orientation by turning the screen of the head-mounted

display black whenever the original heading (i.e., the original

virtual facing direction) changed by more than 10°.

Of the 13 locations in the model, 5 were initial locations,

and all were target locations. Participants faced each of the 5

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from

122 Frankenstein et al.

a b

c d

Fig. 1. Illustration of the setup for the pointing task. Participants saw a virtual model of Tübingen, Germany, a snapshot from which is illustrated here (a). To perform the task (b), participants wore a head-mounted display and used a joystick to point. On each trial, participants began the task in one of 5 initial locations and pointed toward 3 target locations. Across trials, participants were oriented in 12 directions in each of the initial virtual locations (c); some of these orientations were aligned with a street in the model (150° and 330° in the illustration here). There were 13 virtual locations in the model; 5 locations served as both initial locations and target locations (d; illustrated here with Os), and the remaining 8 locations served only as target locations (illustrated here with Xs).

initial locations from 12 orientations (differing by intervals of

30°; see Fig. 1), for a total of 60 trials and 180 pointing

responses per participant. Trials were fully randomized, with

the constraint that all targets were pointed to equally often

and no target was pointed to twice in one trial. Intertrial

intervals were determined by participants. We recorded self-

localization time, pointing latency, and absolute pointing error.

After completing all pointing trials, participants were asked to

draw a map that included all of the locations they had pointed

to in the experiment.

Only participants who had been able to identify all of the

target locations (from photographs) and all of the initial loca-

tions (from 360° snapshots) before the experiment were

included in our analysis. Participants received written and oral

instructions that indicated the exact spot they should point to

for each target (e.g., they were instructed to point toward the

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from

The Orientation of Mental Maps 123

main entrance when pointing toward the university building

and to point toward a tower when pointing toward the fire sta-

tion). Participants were familiarized with the procedure during

training trials that used an initial virtual location not used in

the experiment.

We examined pointing performance as a function of local

orientation and global orientation of the participant. Local ori-

entation was defined as the minimal angle between street ori-

entation and head orientation, ranging from −90° to 90° and

categorized in increments of 30°. Global orientation was

defined as head orientation relative to either a north-facing ori-

entation (for a map-based oriented GRF) or the orientation of

the map the participant had drawn (for individual oriented

GRFs) and was also categorized in increments of 30°. To test

whether performance decreased as misalignment increased,

we conducted three contrast analyses, one with contrasts cen-

tered on a local orientation and the other two with contrasts

centered on the global orientations (see Keppel & Wickens,

2004, for details). The contrast weights for the LRF analysis

were 9/2/–5/–12/–5/2/9, with –12 corresponding to a street-

aligned orientation; the contrast weights for the GRF analyses

were 3/2/1/0/–1/–2/–3/–2/–1/0/1/2, with –3 corresponding to a

north-facing orientation or the individual map orientation. To

estimate distance effects, we correlated pointing accuracy and

latency with the euclidean distance between the initial location

and the target. Pointing accuracy for each participant was bet-

ter than the chance level of 90° that would result from random

pointing, ts(179) < −3.86, ps < .001. Values that deviated from

the overall mean by more than 2 standard deviations (< 4% of

pointing responses) were eliminated from our statistical

analysis.

Results

Global reference frames acquired from maps

If participants acquired an oriented GRF from a map, their per-

formance would be best for north-facing head orientations, and

there would be no distance effect. Our results were consistent

with this pattern. Participants’ average pointing accuracy var-

ied as a function of their global head orientation (Fig. 2a),

F(5.9, 153.9) = 66.29, p < .001, ηp

2 = .72 (Greenhouse-Geisser

corrected). Global head orientation had no effect on self-

localization time, F(11, 286) = 1.59, p = .103, ηp

2 = .06, or

pointing latency (F < 1). Each participant’s pointing accuracy

was predicted by the map-based-GRF contrast, and although

participants’ performance was relatively improved in south-

facing orientations, accuracy decreased linearly with the angle

of misalignment, smallest t(163) = 2.27, all ps < .026. This lat-

ter finding suggests that participants used mental rotation to

compensate for misalignments.

Neither the correlation between target distance and point-

ing error, rs = −.32–.24 (M = −.01, SD = .14; see Fig. 2b),

nor the correlation between target distance and pointing

latency, rs = −.35–.24 (M = −.05, SD = .16; see Fig. 2c), was

significantly different from 0, t(26) = −0.42, p = .679, and

Fig. 2. Pointing error as a function of global head orientation (a) and correlations between pointing performance and target distance (b, c). In (a), the gray lines show results for individual participants, and the black line represents averaged pointing error (error bars represent standard errors). The histograms show the distribution of individual correlations (b) between pointing error and the euclidean distance to a target and (c) between pointing latency and the euclidean distance to a target. The black vertical lines in the histograms correspond to the absence of a distance effect (r = .0); the dashed gray lines correspond to an assumed small distance effect (r = .20).

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from

124 Frankenstein et al.

t(26) = −1.55, p = .132, respectively. Analyses at the individ-

ual level revealed a small negative correlation between target

distance and pointing error for 2 participants (strongest corre-

lation: r = −.32, p < .014) and a small negative correlation

between target distance and pointing latency for 5 participants

(strongest correlation: r = −.35, p < .006). The data are consis-

tent with all predictions from a map-based-GRF account: Par-

ticipants’ performance was best when they were oriented

north, and there was no positive correlation between perfor-

mance (magnitude of error, latency) and distance.

Global reference frames acquired

from navigation

The orientation of GRFs acquired from navigation is likely to

differ between participants. We used the maps participants had

drawn to estimate the orientation of individual GRFs. Two

independent experimenters rated the orientations of the maps

as north oriented, east oriented, south oriented, or west ori-

ented; the raters agreed in their judgments of 26 of the 27

maps (interrater reliability: κ = .93), and we included the 26

participants who had drawn these maps in our analyses. Most

participants drew south-oriented maps (17 south-oriented, 5

north-oriented, and 4 west-oriented maps). Pointing accuracy

differed as a function of the relation between head orientation

and map orientation, F(2.7, 67.7) = 5.62, p = .002, ηp

2 = .18

(Fs < 1 for self-localization and pointing time). All individual

contrast analyses for participants with south-oriented maps

were significant but revealed inverse values indicating that

performance was best when these participants were oriented

north; participants who drew west-oriented maps also pointed

most accurately when they were facing north. Contrast analy-

ses were not significant for any participants with west-oriented

maps (all ps > .136, largest t = 1.50). If participants had relied

on oriented GRFs acquired from navigation, a positive corre-

lation between target distance and pointing error would be

expected. However, even an assumed small distance effect

(r = .20; see Fig. 2) is inconsistent with the observed data—

error magnitude: t(26) = −7.64, p < .001; latency: t(26) =

−7.93, p < .001. Our results are therefore inconsistent with the

use of navigation-based oriented GRFs.

Local reference frames

The LRF account predicts better performance for body orien-

tations aligned, rather than misaligned, with a local street.

Indeed, pointing accuracy varied as a function of alignment

between body orientation and street orientation, F(4.08,

106.09) = 12.25, p < .001, ηp

2 = .32. However, the observed

effect contradicted this hypothesis: Six participants’ perfor-

mance was worse when they were aligned with a street,

t(169) < –2.06, ps < .041; only 1 participant’s performance

was better in this condition, t(171) = 2.62, p = .010. There was

no effect of street alignment on self-localization time, F(3.80,

98.91) = 2.29, p = .068, ηp

2 = .08, or pointing latency (F < 1).

Distance effects for pointing latencies and error magnitude

were not observed. Our data are thus inconsistent with the pre-

dictions of the LRF account.

Discussion

The fact that a north-facing orientation benefitted performance

and the fact that we found no distance effects suggest that par-

ticipants used representations based on city maps, which are

north-oriented, single-frame representations. Error magnitude

increased as participants’ misalignment from a north-facing

orientation increased; this finding suggests that participants

used mental rotation to compensate for misalignment (Iachini

& Logie, 2003). However, participants’ performance in

contra-aligned (i.e., south-facing) body orientations was better

than would be expected if participants were using mental rota-

tion. This pattern of results is similar to results found in prior

research on map-based memory (Hintzman, O’Dell, & Arndt,

1981) and suggests that participants applied a strategy other

than mental rotation in contra-aligned orientations.

The pattern of increasing errors with increasing misalign-

ment from a north-facing orientation held for every partici-

pant, which indicates a lack of variation in strategy. Tübingen

is not organized on a north-south grid and has no widely visi-

ble characteristic landmarks on a north-south axis. Moreover,

participants were not physically oriented north during data

collection. Alternative explanations based on these factors are

therefore implausible. These alternative explanations also

would not explain the lack of a distance effect.

Our data do not support accounts of individual global

or local reference frames acquired by navigation (McNamara

et al., 2008; Meilinger 2008; O’Keefe, 1991; Poucet, 1993;

Trullier et al., 1997; Wang & Spelke, 2002) or accounts of

orientation-free memory for highly familiar environments

experienced from multiple perspectives (Byrne et al., 2007;

Evans & Pezdek, 1980; Gallistel, 1990; Sholl, 1987). How-

ever, long-term experience navigating a city without the use of

maps might yield orientation-free representations as well as

individual GRFs or LRFs. Compared with our study, most pre-

vious studies that appear to support individual GRFs or LRFs

have used smaller scale spaces, shorter learning periods with

less perceptual input (e.g., only visual input), or different tasks

(e.g., route navigation). Indeed, altering these factors in our

experimental paradigm might provide evidence for different

representations, as might testing populations without experi-

ence using maps (e.g., children) or populations with experi-

ence using different kinds of maps (e.g., Japanese people).

Our finding that Western participants’ memory for a highly

familiar city seemed to rely strongly on maps, even though par-

ticipants had spent significantly more time navigating the city

than looking at maps of it, is surprising. Some participants

reported not having looked at a map of Tübingen for decades.

Unlike navigation, which provides rich multimodal experiences,

maps are mainly perceived visually, and the visual information

they provide is limited. Maps typically display few visual

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from

The Orientation of Mental Maps 125

features of a location (e.g., the geometry) if they display any

visual features at all, and few locations are explicitly highlighted

in most maps of cities; most of the locations used in our experi-

ment were not highlighted in maps of Tübingen. Participants

therefore had to identify their location and orientation using

knowledge they had gained from navigating the city and then

relate this navigational knowledge to map knowledge, thereby

switching from a ground perspective to a bird’s-eye view.

Why did participants make this effort and weight map-

based representational structures more heavily than represen-

tations derived from multisensory navigational experience?

Maps represent an environment within a single reference

frame and accurately reflect multiple spatial relations that do

not need to be verified by and adapted to accord with further

navigational experience. They present a reliable structure for

organizing complex navigational experiences and contain sur-

vey relations required for pointing. Remembering and men-

tally rotating a city map might be computationally easier than

deriving survey relations by integrating multiple navigational

experiences within a single reference frame. Our results sup-

port the popular belief that people have access to something

like a map in their heads and suggest that—at least for our

participants from Tübingen—this map is oriented north.

Acknowledgments

The authors thank Michael Weyel, Stephan Streuber, Hans-Günther

Nusseck, and Sally Linkenauger for their help and the land surveying

office of Tübingen, Germany, for providing detailed maps.

Declaration of Conflicting Interests

The authors declared that they had no conflicts of interest with

respect to their authorship or the publication of this article.

Funding

This research was supported by the Max Planck Society and the

German Research Foundation (Grant ME 3476/2 “Survey Knowl-

edge” and Grant SFB/TR8 “Spatial Cognition”) and by the National

Research Foundation of Korea’s World Class University program

(Grant R31-10008).

References

Byrne, P., Becker, S., & Burgess, N. (2007). Remembering the past

and imagining the future: A neural model of spatial memory and

imagery. Psychological Review, 114, 340–375.

Christou, C., & Bülthoff, H. H. (1999). View dependence in scene

recognition after active learning. Memory & Cognition, 27, 996–

1007.

Evans, G. W., & Pezdek, K. (1980). Cognitive mapping: Knowledge

of real-world distance and location information. Journal of Experi-

mental Psychology: Learning, Memory, and Cognition, 6, 13–24.

Gallistel, C. R. (1990). The organization of learning. Cambridge,

MA: MIT Press.

Gillner, S., Weiß, A. M., & Mallot, H. A. (2008). Visual homing in

the absence of feature-based landmark information. Cognition,

9, 105–122.

Hintzman, D. L., O’Dell, C. S., & Arndt, D. R. (1981). Orientation in

cognitive maps. Cognitive Psychology, 13, 149–206.

Iachini, T., & Logie, R. H. (2003). The role of perspective in locating

position in a real-world, unfamiliar environment. Applied Cogni-

tive Psychology, 17, 715–732.

Keppel, G., & Wickens, T. D. (2004). Design and analysis: A

researcher’s handbook (4th ed.). Upper Saddle River, NJ: Pren-

tice Hall.

Levine, M., Marchon, I., & Hanley, G. (1982). The placement and

misplacement of you-are-here maps. Environment & Behavior,

16, 139–157.

Loomis, J. M., Klatzky, R. L., Golledge, R. G., Cicinelli, J. G.,

Pellegrino, J. W., & Fry, P. A. (1993). Nonvisual navigation by

blind and sighted: Assessment of path integration ability. Journal

of Experimental Psychology: General, 122, 73–91.

McNamara, T. P., Sluzenski, J., & Rump, B. (2008). Human spatial

memory and navigation. In J. Byrne (Ed.), Learning and mem-

ory: A comprehensive reference (Vol. 2, pp. 157–178). Oxford,

England: Elsevier.

Meilinger, T. (2008). The network of reference frames theory: A

synthesis of graphs and cognitive maps. In C. Freksa, N. S.

Newcombe, P. Gärdenfors, & S. Wölfl (Eds.), Spatial cognition

(Vol. 6, pp. 344–360). Berlin, Germany: Springer.

Meilinger, T., Knauff, M., & Bülthoff, H. H. (2008). Working mem-

ory in wayfinding: A dual task experiment in a virtual city. Cogni-

tive Science, 32, 755–770.

O’Keefe, J. (1991). An allocentric spatial model for the hippocampal

cognitive map. Hippocampus, 1, 230–235.

Poucet, B. (1993). Spatial cognitive maps in animals: New hypoth-

eses on their structure and neural mechanisms. Psychological

Review, 100, 163–182.

Richardson, A. E., Montello, D., & Hegarty, M. (1999). Spatial

knowledge acquisition from maps and from navigation in real

and virtual environments. Memory & Cognition, 27, 741–750.

Sholl, M. J. (1987). Cognitive maps as orienting schemata. Journal

of Experimental Psychology: Learning, Memory, and Cognition,

13, 615–628.

Sun, H.-J., Chan, G. S. W., & Campos, J. L. (2004). Active naviga-

tion and orientation-free spatial representations. Memory & Cog-

nition, 32, 51–71.

Thorndyke, P. W., & Hayes-Roth, B. (1982). Differences in spatial

knowledge acquired from maps and navigation. Cognitive Psy-

chology, 14, 560–589.

Trullier, O., Wiener, S. I., Berthoz, A., & Meyer, J.-A. (1997). Bio-

logically based artificial navigation systems: Review and pros-

pects. Progress in Neurobiology, 51, 483–544.

Wang, F. R., & Spelke, E. S. (2002). Human spatial representa-

tion: Insights from animals. Trends in Cognitive Sciences, 6,

376–382.

at Max Planck Society on April 11, 2012pss.sagepub.comDownloaded from