Cogn Process (2015) 16 (Suppl 1):S437–S441 DOI 10.1007/s10339-015-0726-5

SHORT REPORT

Configurational salience of landmarks: an analysis of sketch maps using Space Syntax Rul von Stu¨lpnagel1 • Julia Frankenstein2

Published online: 4 August 2015 Ó Marta Olivetti Belardinelli and Springer-Verlag Berlin Heidelberg 2015

Abstract We conducted a visibility graph analysis (a Space Syntax method) of a virtual environment to examine how the configurational salience of global and local landmarks (i.e., their relative positions in the environment) as compared to their visual salience affects the probability of their depiction on sketch maps. Participants of two experimental conditions produced sketch maps from memory after exploration with a layout map or without a map, respectively. Participants of a third condition produced sketch maps in parallel to exploration. More detailed sketch maps were produced in the third condition, but landmarks with higher configurational salience were depicted more frequently across all experimental conditions. Whereas the inclusion of global landmarks onto sketch maps was best predicted by their size, both visual salience and isovist size (i.e., the area a landmark was visible from) predicted the frequency of depiction for local landmarks. Our findings imply that people determine the relevance of landmarks not only by their visual, but even more by their configurational salience. Keywords Landmark  Landmark salience  Sketch map  Space Syntax

& Rul von Stu¨lpnagel [email protected] Julia Frankenstein [email protected] 1

Center for Cognitive Science, University of Freiburg, Friedrichstr. 50, 79098 Freiburg, Germany

2

Lehrstuhl fu¨r Bildungspsychologie, Universita¨t Mannheim, Postfach 103462, 68131 Mannheim, Germany

Introduction When people get to know an unknown area (e.g., a new neighborhood), they frequently integrate their navigational experiences into a cognitive map, which can be assessed by sketch mapping (Ishikawa and Montello 2006). Landmarks are one central element of such human navigation and orientation processes. A number of approaches aim to determine the relevance of a landmark in relation to a specific route (Michon and Denis 2001), or in a local context as a function of its visual, semantic, and structural salience (Raubal and Winter 2002). Whereas some approaches scrutinize visual salience, few take the structural and configurational properties of an environment into account (Ro¨ser et al. 2012). Space Syntax is an approach that encompasses a set of methods geared toward transferring the configurational properties of an environment into a quantifiable graph (Hillier 2008). It is thus suitable for examining the configurational salience of landmarks as assessed with sketch maps. For example, Haq and Girotto (2003) showed that the configurational properties of a network of corridors are reflected in sketch maps of the original building. Rafailaki (2006, chapter 4.3) found that the frequency with which landmarks were depicted by people asked to draw a sketch map of their home city correlated with the relative centrality of those landmarks’ location in the environmental graph. Krukar and Dalton (2013) examined the relationship between gaze patterns and landmarks’ configurational properties using a visibility graph analysis (VGA). In a VGA an even grid of cells is applied to all accessible spaces of an environment. Connectivity, which, in a VGA, refers to the number of visible cells from a specific cell (or set of cells), is one central measure in Space Syntax. Together these visible cells form an isovist whose

123

S438

properties (e.g., its size) have strong ties to human orientation behavior (Conroy-Dalton 2001, chapter 8). Another important measure is integration, derived from the average visual distance of a cell to all other cells in the environment. Using these measures, we examined the impact of a landmark’s configurational salience as compared to its visual salience on the probability this landmark is depicted on sketch maps of a to-be-explored environment. In our study, we included two established exploration conditions: One featuring a layout map (MAP) and one without additional aids (FREE), thus accounting for potentially different environmental representations (Shelton and McNamara 2004). However, the probability of a landmark being depicted on a map sketched from memory could be affected by exposition (i.e., how often a landmark was passed or spotted during exploration). Thus, sketch maps were produced in parallel to exploration in a third condition (SKETCH), where we expect that maps are sketched procedurally, and the inclusion of a landmark reflects an intuitive estimation of its relevance for future orientation.

Methods Forty-five students (44 % males; age: M = 22, SD = 2) participated and were individually tested and randomly assigned to the experimental conditions SKETCH, MAP, or FREE. Participants were instructed to explore a virtual environment resembling an urban scene presented on three 2400 monitors arranged in a semi-circle and make themselves as familiar as possible with it (see Fig. 1).1 Movement was controlled using a gamepad. Participants in the MAP condition were provided with a layout map not depicting local landmarks and buildings’ interiors during exploration. Participants in the FREE condition explored the environment without additional aids. Participants in both conditions explored the environment for 8 min before producing a sketch map from memory for 2 min. Participants in the SKETCH condition were instructed to produce a sketch map in parallel to exploration for 10 min. All participants were encouraged to sketch all environmental elements they considered relevant for later orientation (see Fig. 1). The environment included 20 global landmarks (i.e., larger geographic features such as streets and buildings) and 48 local landmarks (e.g., cars). Sketch maps from all participants were analyzed for those landmarks. As an indication of configurational salience we ran a VGA using 1

We analyzed the proportion explored in the three experimental conditions, extracted from recordings of the exploration phase. Less of the environment was explored in the SKETCH condition (M = 68 %, SD = 13) than in the MAP condition (M = 80 %, SD = 9), t(24) = -2.65, p \ .02. The FREE condition (M = 76 %, SD = 9) did not differ from the other conditions, both p [ .09.

123

Cogn Process (2015) 16 (Suppl 1):S437–S441

DepthmapX (Varoudis 2012) to determine landmark size (i.e., the number of cells a landmark occupied), isovist size (i.e., the number of cells a landmark was completely or partially visible from) and integration (i.e., the average visual distance to all cells in the environment, see Fig. 1). In order to explore the relation between configurational salience and visual salience, visual salience of all landmarks was estimated by five independent raters on 5-point Likert scales.2

Results First we examined the sketch maps per experimental condition. All descriptive and corresponding statistical values are reported in Table 1. Participants of the SKETCH condition produced sketch maps featuring significantly more global as well as local landmarks than both other conditions. Corresponding differences were also found for the configurational properties of global landmarks: size, isovist size, and integration of the depicted landmarks in the SKETCH condition were consistently smaller than in both other conditions. However, one-sample t tests showed that in all three conditions participants preferred to sketch global landmarks larger, more visible, and more integrated than the respective average.3 In contrast, visual salience of the depicted global landmarks neither differed between the conditions nor from the average. Findings were similar but less pronounced for local landmarks: Participants in the SKETCH condition included less visible and less integrated landmarks as compared to participants in the MAP condition. Although participants of all three conditions preferred more visible and more integrated local landmarks for their sketch maps, their size was above average only for the SKETCH condition. Visual salience of the depicted local landmarks did not differ between the conditions, but was consistently above the average. Second we examined the sketch maps per landmark.4 We computed multiple linear regression models for global and local landmarks, respectively (see Fig. 2). We included visual salience, landmark size, isovist size, and integration; and used a stepwise selection method to determine the best 2

Raters estimated how visually obtrusive a landmark was in comparison to the immediate surroundings, but irrespective of its size. Inter-rater reliability was satisfying, ICC(2,5) = .87. Thus, mean values were computed and used in all analyses reported below. 3 Isovist size of global and local landmarks did not differ from the mean for the SKETCH condition. We attribute this effect to a regression towards the mean due to the significantly higher number of depicted landmarks in the SKETCH condition as compared to the other conditions. 4 Separate analyses for the experimental conditions revealed highly similar patterns. Thus, the data were collapsed over this factor for the analyses reported below.

Cogn Process (2015) 16 (Suppl 1):S437–S441

S439

Fig. 1 Left VGA of the experimental environment indicating integration (red = high; blue = low). The white rectangle exemplifies the size of an exemplary landmark. The dotted line approximates this

landmark’s isovist size. Top right view of the experimental environment with roofs of accessible buildings being removed for illustration reasons. Bottom right example of a sketch map (color figure online)

Table 1 Descriptive means (and standard deviations) of all dependent variables, separately for each experimental condition Experimental condition: dependent variable Global landmarks (maximum 20)

SKETCH 9.25 (2.62)

MAP 6.31 (3.60)

FREE 6.69 (2.36)

Effect between conditions (significant LSD post hoc tests) F(2,42) = 4.52, g2p = .18* (SKETCH vs. MAP**; SKETCH vs. FREE*)

Visual salience (M = 1.53, SD = 0.39)

1.58 (.11)

1.55 (.17)

1.58 (.14)

F\1

Landmark size (M = 264, SD = 109)

344* (53)

443* (171)

476* (160)

F(2,42) = 3.78, g2p = .15* (SKETCH vs. MAP*; SKETCH vs. FREE*)

Isovist size (M = 2558, SD = 482)

2703 (534)

3397* (979)

3376* (851)

F(2,42) = 3.71, g2p = .15* (SKETCH vs. MAP*; SKETCH vs. FREE*)

Integration (M = 7.70, SD = .40)

7.87* (.29)

8.51* (.89)

8.39* (.76)

F(2,42) = 3.77, g2p = .15* (SKETCH vs. MAP*; SKETCH vs. FREE*)

10.39 (4.33)

F(2,42) = 22.35, g2p = .52*** (SKETCH vs. MAP***; SKETCH vs. FREE***)

Local landmarks (maximum 48)

21.25 (8.29)

8.13 (3.65)

2.55* (.21)

2.66* (.45)

2.69* (.32)

F\1

Landmark size (M = 13, SD = 14)

14.37* (2.28)

12.93 (1.91)

15.23 (5.51)

F = 1.67, p [ .20

Isovist size (M = 1508, SD = 855)

1590 (236)

1857* (304)

1806* (397)

F(2,42) = 3.24, g2p = .13* (SKETCH vs. MAP*)

Integration (M = 8.90, SD = 2.24)

8.89* (.43)

9.83* (1.00)

9.35* (.88)

F(2,42) = 5.43, g2p = .21* (SKETCH vs. MAP*)

Visual salience (M = 2.39, SD = .95)

Indications of statistical significance in the columns SKETCH, MAP, and FREE refer to one-sample t tests against the mean presented in the first column. The right column presents the results of (SKETCH vs. MAP vs. FREE) ANOVAs and all significant LSD post hoc test effects below * p \ .05, ** p \ .01, *** p \ .001

predictors for the criterion of a landmark’s depiction frequency. For global landmarks, the only significant predictor proved to be landmark size, F(1,18) = 20.81, p \ .001. In contrast, the model for local landmarks indicated two significant predictors, visual salience and isovist size, F(2,45) = 6.76, p \ .01.

Discussion We examined how a landmark’s configurational salience (i.e., its relative location in the environment) as compared to its visual salience affects the frequency with which this landmark is depicted on sketch maps produced both from

123

S440

Cogn Process (2015) 16 (Suppl 1):S437–S441

Fig. 2 Stepwise multiple linear regression models testing visual salience, landmark size, isovist size and integration as predictors for the depiction frequency of global landmarks (left panel, R2 = .54) and local landmarks (right panel, R2 = .14 for Step 1, DR2 = .10 for Step 2 (p \ .03). *p \ .05

memory and in parallel to exploration. In general, all participants preferred to sketch landmarks that were larger, more visible, and more integrated than average. Multiple regression analyses indicated that the best predictor for the depiction of a global landmark was its size. In contrast, the inclusion of local landmarks was predicted by both visual salience and isovist size (i.e., a landmark’s visibility). This implies that people estimate the configurational salience of a landmark due to its function in the surrounding environment and consider this factor about as much or more important than this landmark’s visual salience. Surprisingly, integration (a measure which has repeatedly been found to predict human orientation behavior, e.g., Conroy-Dalton 2001; Haq and Girotto 2003) failed to explain significant amounts of variance. This may be a result of the deliberately unspecific instructions participants received prior to exploration. Provided a task emphasizing a global understanding of the environment (e.g., von Stu¨lpnagel et al. 2014), we would expect the importance of integration to increase. The SKETCH condition produced sketch maps featuring significantly more global as well as local landmarks than the conditions MAP and FREE, who had to retrieve all spatial information they depicted on their sketch maps from memory. The finding that the SKETCH condition depicted landmarks of lesser configurational salience than the MAP and the FREE condition results most likely from this difference in the number of landmarks sketched (i.e., representing a regression toward the mean). We conclude that sketch mapping in parallel to exploration differs from sketch mapping from memory quantitatively (i.e., being more fine-grained) rather than qualitatively (i.e., emphasizing different landmark properties). We are not aware of experimental investigations of sketch mapping in parallel to exploration, but this approach resembles active navigation, which can benefit the processing of spatial information (von Stu¨lpnagel and Steffens 2012). Future research should compare sketch maps produced in parallel to exploration and from memory by the same participant, thus scrutinizing how configurational salience affects the shift from a landmark’s initial perception to its memorization.

123

Taken together, our results imply that people choose landmarks for future orientation not only due to their visual salience, but also strategically with regard to their configurational properties in the surrounding environment. Acknowledgments This research was granted by the SFB/TR8— Spatial Cognition. We thank Sascha Poppitz for programming the environment; Sascha Crede, Saskia Leymann, and Nicolas Holland for collecting and processing the data; and Tyler Thrash, Kai Preuß, and Stephanie Schwenke for valuable input during the preparation of this manuscript.

References Conroy-Dalton R (2001) Spatial navigation in immersive virtual environments. University of London, London Haq S, Girotto S (2003) Ability and intelligibility: wayfinding and environmental cognition in the designed. In: Presented at the 4th international space syntax symposium, London Hillier B (2008) The social logic of space (Reprint). Cambridge University Press, Cambridge, MA Ishikawa T, Montello D (2006) Spatial knowledge acquisition from direct experience in the environment: individual differences in the development of metric knowledge and the integration of separately learned places. Cogn Psychol 52(2):93–129. doi:10. 1016/j.cogpsych.2005.08.003 Krukar J, Dalton RC (2013) Spatial predictors of eye movement in a gallery setting. In: Kiefer P, Giannopoulos I, Raubal M, Hegarty M (eds) Eye tracking for spatial research, Proceedings of the 1st international workshop (in conjunction with COSIT 2013), pp 14–19 Michon P-E, Denis M (2001) When and why are visual landmarks used in giving directions? In: Montello D (ed) Spatial information theory. Springer, Berlin, pp 292–305 Rafailaki E (2006) The implications of the ‘‘palimpsest’’ of the grids of the main city of Piraeus on creation, transmission and application of cognitive knowledge. UCL (University College London), London Raubal M, Winter S (2002) Enriching wayfinding instructions with local landmarks. In: Proceedings of the second international conference on geographic information science. Springer, London, pp 243–259 Ro¨ser F, Hamburger K, Krumnack A, Knauff M (2012) The structural salience of landmarks: results from an on-line study and a virtual environment experiment. J Spat Sci 57(1):37–50. doi:10.1080/ 14498596.2012.686362 Shelton AL, McNamara TP (2004) Orientation and perspective dependence in route and survey learning. J Exp Psychol Learn Mem Cogn 30(1):158–170

Cogn Process (2015) 16 (Suppl 1):S437–S441 Varoudis T (2012) DepthmapX multi-platform spatial network analysis software (version 0.30). http://varoudis.github.io/ depthmapX/ Von Stu¨lpnagel R, Steffens MC (2012) Can active navigation be as good as driving? A comparison of spatial memory in drivers and backseat drivers. J Exp Psychol Appl 18(2):162–177. doi:10. 1037/a0027133

S441 Von Stu¨lpnagel R, Kuliga S, Bu¨chner SJ, Ho¨lscher C (2014) Supraindividual consistencies in navigator-driven landmark placement for spatial learning. In: Bello P, Guarini M, McShane M, Scassellati B (eds) Proceedings of the 36th annual conference of the cognitive science society. Cognitive Science Society, Austin, TX, pp 1706–1711

123