destination descriptions in urban environments phd … · as a true guru, stephan led me through...

Department of Geomatics

The University of Melbourne

DESTINATION DESCRIPTIONS

IN URBAN ENVIRONMENTS

PhD THESIS

Martin Tomko

Submitted in total fulfillment of the requirementsof the degree of Doctor of Philosophy

April, 2007

Abstract

An important difference exists between the way humans communicate routeknowledge and the turn-by-turn route directions provided by the majority ofcurrent navigation services. Navigation services present route directions withthe same amount of detail regardless the route segment’s significance in the in-structions, user’s distance from the destination, and finally the level of user’sfamiliarity with particular parts of the environment.

A significant feature of human-generated route directions provided to peo-ple is the hierarchical communication of route knowledge. References are madeto a simplified structure of the environment. Communication partners exchangeroute directions assuming a shared knowledge of the coarse environment’s struc-ture. Such destination descriptions provide an increased amount of detail as thedescription approaches the proximity of the destination of the route.

The research presented in this thesis aims to improve the communication ofnavigation information by presenting a formal model enabling the selection ofreferences for destination descriptions. The model is based on the analysis of thereflection of the structure of the urban environment in destination descriptionsprovided by locals. In such spatial communication, common knowledge of thecoarse structure of the city is inferred.

The main contribution of this thesis is the analysis of the reflection of thestructure of an urban environment in the route directions exchanged betweenpeople with at least coarse knowledge of the environment, and the formalizationof these principles in a computational model that enables automated selectionof referents for destination descriptions. In the approach presented, the environ-mental elements of the city structure are hierarchically integrated together witha model of the communication processes underlying the creation of destinationdescriptions .

Automated creation of directions with a variable level of detail will improvethe ability to reflect the alteration of local conditions. The resulting route di-rections are usually shorter than those created by current navigation services,and thus lower the cognitive workload of the wayfinder. The benefactors ofsuch a system are wayfinders frequently traveling to unfamiliar destinations inpartially-known urban environments, such as the police, emergency managementand tourism services, but also locals—everyday users of Web based navigationportals.

iii

Keywords

destination descriptions, wayfinding, pragmatic communication, relevance, spa-tial knowledge, a-priori spatial knowledge, familiarity

iv

Declaration

This is to certify that:

(i) the thesis comprises only my original work towards the PhD except whereindicated,

(ii) due acknowledgment has been made in the text to all other material used,

(iii) the thesis is less than 100,000 words in length, exclusive of table, maps,bibliographies, appendices and footnotes.

Martin Tomko

v

Preface

The work presented in this thesis has been supported by the Cooperative ResearchCentre for Spatial Information, whose activities are funded by the AustralianCommonwealth’s Cooperative Research Centres Programme.

The street network dataset of Hannover is part of the ATKIS Basis DLMdataset provided by the National Mapping Agency of Lower Saxony (Landesver-messung und Geobasisinformation Niedersachsen).

vii

Publications

During the course of this project, a number of publications and public presenta-tions have been made which are based on the work presented in this thesis. Theyare listed here for reference.

Journal Papers

Tomko, M., Winter, S., Claramunt, C., to appear 2007. Expe-riential Hierarchies of Streets. Computers, Environment andUrban Systems

Winter, S., Tomko, M., Elias, B., Sester, M., to appear 2007.Landmark Hierarchies in Context. Environment and PlanningB: Planning and Design

Tomko, M., Winter, S., 2006c. Recursive Construction of Gran-ular Route Directions. Journal of Spatial Science 51 (1), 101–115

Book Chapters

Tomko, M., Winter, S., 2006b. Initial Entity Identification forGranular Route Directions. In: Kainz, W., Riedl, A., Elmes,G. (Eds.), Progress in Spatial Data Handling. 12th Interna-tional Symposium on Spatial Data Handling. Springer-Verlag,Vienna, Austria, pp. 43–60

Winter, S., Tomko, M., 2006. Translating the Web Semanticsof Georeferences. In: Taniar, D., Wenny Rahayu, J. (Eds.),Web Semantics and Ontology. Idea Group Publishing, Her-shey, Pennsylvania, USA, pp. 297–333

Reviewed Conference Papers

Tomko, M., Winter, S., 2005. Reconstruction of Scenes fromGeo-Referenced Web Resources. In: Spatial Science Confer-ence. Spatial Science Institute, Melbourne, Australia

Tomko, M., 2004b. Case Study - Assessing Spatial Distribu-tion of Web Resources for Navigation Services. In: Claramunt,C., Boujou, A., Kwon, Y. J. (Eds.), 4th International Work-shop on Web and Wireless Geographical Information SystemsW2GIS 2004. Goyang, Korea

ix

Invited Talk Winter, S., Tomko, M., 2005. Route Directions of VaryingGranularity. In: Seminar on Spatial Cognition. Schloss Dagstuhl,Germany

Abstract Reviewed Papers

Tomko, M., Winter, S., 2006a. Considerations for EfficientCommunication of Route Directions. In: Cartwright, W., Yoshida,H., Andrienko, G. (Eds.), ISPRS Workshop on Spatial DataCommunication and Visualization. ISPRS, Vienna, Austria

Winter, S., Tomko, M., 2004. Shifting the Focus in MobileMaps. In: Morita, T. (Ed.), Joint Workshop on Ubiquitous,Pervasive and Internet Mapping UPIMap2004. Tokyo, Japan

Technical Reports

Winter, S., Klippel, A., Tomko, M., May 2, 2005 2005. Deliver-able 3.3/2: Experiments on Usability. Internal report, CRC-SIAustralia

Tomko, M., August 23, 2004 2004a. Analysis of WayfindingScenarios. Internal project report, CRC-SI Australia

x

Acknowledgements

The three years during which I worked on this thesis were an amazing experiencewhich would not have been possible without the contribution of many people.First and foremost, I would like to express my gratitude to my supervisor, Dr.Stephan Winter. As a true guru, Stephan led me through the obstacles of sci-entific thinking and writing on the way to become a researcher. Without hisencouragement, guidance, support and patience, this thesis would not exist.

My sincere thanks are extended also to my second supervisor, Dr. MattDuckham for all the feedback and suggestions, and to Dr. Alexander Klippelfor the sometimes turbulent discussions. Without the pressure to answer Alex’squestions, many details in this thesis would remain unsolved. I would also liketo thank my industry mentor Maurits van der Vlugt for advice throughout thisproject.

I express my gratitude to Prof. Christophe Claramunt of the Naval AcademyResearch Center as well as to Dr. Monika Sester and Dr. Birgit Elias from theUniversity of Hannover for inspiring collaboration on parts of this thesis and forproviding the landmark dataset. Furthermore, Birgit was always keen to help withher local knowledge of Hannover and assess the results produced in this thesis.My thanks are extended to the colleagues from the Department of Geomatics,Stefanie Andrae, Jochen Wilneff, Joanne Poon, Stephan Hansen, Sue Hope, JaneInall, Zaffar Sadiq and Patrik Laube for discussions, support, encouragement, funand table tennis matches, and especially Anna Boin for reading through the draftof this thesis.

Writing a PhD thesis can be a lonely experience, the more that I wrotemine far from home. First of all, I would like to thank Callum Eastwood forsupport and friendship which I believe will last a long way beyond the PhDyears. Cheers, Bro! Ned Rogers, Marina Carpinelli, Elisa Toulson and MichaelLaity, along with the fellow paddlers, climbers and bushwalkers of the MelbourneUniversity Mountaineering Club created my closest social network. I would liketo thank Ertan Yesilnacar for being an understanding and patient flatmate duringthe whole duration of this research.

Finally, I would like to express my deepest gratitude to a few very specialpeople - my parents, Jana and Jan Tomko, and my brother Jakub Tomko, for theyears of love and encouragement. They always trusted me, and I always tried tomake them proud of being my parents. Last, but not least, I would like to thankmy partner Miranda Smith for her patience, support and love during the stressfulperiod of producing this thesis, as well as for proofreading the final drafts of thethesis. Thank you for everything!

xi

Contents

1 Introduction 1

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.2 The Familiar Structure of the Environment and Route Di-rections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.3 Inferential Communication . . . . . . . . . . . . . . . . . . 4

1.1.4 Current Approaches to Route Directions’ Personalization . 5

1.2 Scope and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4 Expected Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.5 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Background 11

2.1 Experiencing Space . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1.1 Mental Representations of Space . . . . . . . . . . . . . . 11

2.1.2 Conceptualization of Space . . . . . . . . . . . . . . . . . . 13

2.1.3 Hierarchical Structure of Spatial Mental Representations . 15

2.1.4 Route Planning and Wayfinding . . . . . . . . . . . . . . . 17

2.2 Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.2.1 Communication Models . . . . . . . . . . . . . . . . . . . 19

2.2.2 The Meaning of Utterances . . . . . . . . . . . . . . . . . 20

2.2.3 Relevance Theory . . . . . . . . . . . . . . . . . . . . . . . 21

2.2.4 Knowledge, Context and Communication . . . . . . . . . . 23

2.2.5 Referential Communication . . . . . . . . . . . . . . . . . 26

2.2.6 Communication about Space . . . . . . . . . . . . . . . . . 27

2.2.7 Directions and Cognitive Ergonomics . . . . . . . . . . . . 29

2.3 Modeling and Formalization . . . . . . . . . . . . . . . . . . . . . 33

2.3.1 Formalization and Functional Programming . . . . . . . . 33

xiii

CONTENTS

2.3.2 Modeling Complex Systems . . . . . . . . . . . . . . . . . 34

2.3.3 Granularity . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.3.4 Spatial Modeling and Network Analysis . . . . . . . . . . . 36

2.3.5 Basic Measures for Network Analysis . . . . . . . . . . . . 38

2.3.6 Basic Elements of the Network . . . . . . . . . . . . . . . 39

3 Destination Descriptions 41

3.1 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.2 Definition of Destination Descriptions . . . . . . . . . . . . . . . . 43

3.3 Structure of Communication . . . . . . . . . . . . . . . . . . . . . 44

3.4 Selection of Referents . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.5 Common Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.6 Characteristics of Destination Descriptions . . . . . . . . . . . . . 53

4 Hierarchical Data Structures 55

4.1 Hierarchical Urban Structures . . . . . . . . . . . . . . . . . . . . 55

4.2 Experiential Hierarchies . . . . . . . . . . . . . . . . . . . . . . . 56

4.3 Composition of the Urban Environment . . . . . . . . . . . . . . . 58

4.4 Hierarchization of Elements of the City . . . . . . . . . . . . . . . 59

4.4.1 Hierarchies of Landmarks . . . . . . . . . . . . . . . . . . 59

4.4.2 Hierarchies of Districts . . . . . . . . . . . . . . . . . . . . 63

4.4.3 Hierarchies of Paths . . . . . . . . . . . . . . . . . . . . . 67

4.5 Integrated Experiential Hierarchy . . . . . . . . . . . . . . . . . . 80

4.6 Concept of Distance in Hierarchical Structures . . . . . . . . . . . 82

5 A Generic Model of Destination Descriptions 83

5.1 Context Specifications . . . . . . . . . . . . . . . . . . . . . . . . 83

5.2 Model Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 85

5.3 Structure of Destination Descriptions . . . . . . . . . . . . . . . . 85

5.4 Relevance of a Reference . . . . . . . . . . . . . . . . . . . . . . . 87

5.5 Rules for Selecting District References . . . . . . . . . . . . . . . 88

5.6 Rules for Selecting Landmarks References . . . . . . . . . . . . . 95

5.7 Rules for Selecting Paths References . . . . . . . . . . . . . . . . 101

5.8 Integrated Destination Descriptions . . . . . . . . . . . . . . . . . 103

6 Model Implementation 105

6.1 Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

xiv

M. TOMKO CONTENTS

6.2 Input and Output Specification . . . . . . . . . . . . . . . . . . . 106

6.3 Main Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6.3.1 Selection of District and Landmark References . . . . . . . 108

6.3.2 Integration of Path References . . . . . . . . . . . . . . . . 109

6.4 Model Verification . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6.4.1 Test of District and Landmark-Based Destination Descrip-tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6.4.2 Test of Destination Descriptions with Paths . . . . . . . . 116

6.5 Observations of the Model Outputs . . . . . . . . . . . . . . . . . 121

7 Conclusions 125

7.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

7.2 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7.3.1 Cognitive Workload and Destination Descriptions . . . . . 129

7.3.2 Reliability of Inference of Common Spatial Knowledge . . 131

7.3.3 Experiential Data Structures . . . . . . . . . . . . . . . . . 132

7.4 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

7.4.1 Descriptions and Directions . . . . . . . . . . . . . . . . . 133

7.4.2 The Where? Question . . . . . . . . . . . . . . . . . . . . 135

7.4.3 Externalization . . . . . . . . . . . . . . . . . . . . . . . . 135

7.4.4 Coupling of Inferential and Agent-Based Systems . . . . . 136

7.4.5 Complex Integrated Experiential Hierarchies . . . . . . . . 136

7.5 Concluding Scenario . . . . . . . . . . . . . . . . . . . . . . . . . 137

A Landmark Names 139

B Input Dataset of Hannover 141

C Program Code 151

D Example Test Cases 157

xv

CONTENTS

xvi

List of Tables

4.1 Number of landmark objects selected by hierarchical level . . . . . 63

4.2 Betweenness centrality vector CBdi

j

of Level 2 districts of central

Hannover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

4.3 Most prominent streets of central Hannover . . . . . . . . . . . . 78

6.1 Patterns in sets of references in destination descriptions. . . . . . 122

A.1 Common names of the landmarks of central Hannover. . . . . . . 139

xvii

LIST OF TABLES

xviii

List of Figures

1.1 Turn-based route directions from Hannover Airport to Luisen-strasse (©2007, Google Maps). . . . . . . . . . . . . . . . . . . . 3

2.1 Acquisition of spatial knowledge. . . . . . . . . . . . . . . . . . . 12

2.2 Wayfinding in unfamiliar environment. . . . . . . . . . . . . . . . 18

2.3 Wayfinding in familiar environment. . . . . . . . . . . . . . . . . . 32

3.1 Structure of navigation instructions . . . . . . . . . . . . . . . . . 44

3.2 Structure of spatial communication for a route with a transitionpoint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.3 Schema of the hierarchy of references in the destination descriptionfor the route from Hannover Airport to Luisenstrasse. . . . . . . . 47

3.4 Mental representations of the speaker (S) and the hearer (H). . . 47

3.5 Detail of the hierarchical structures of the sets S, H and C. . . . 48

3.6 Possible encodings of two contents in messages, and their interpre-tation based on relevance theory. . . . . . . . . . . . . . . . . . . 50

3.7 Detail of the hierarchical structure of the set C. . . . . . . . . . . 50

3.8 Structure of spatial communication with a negotiation dialog. . . 53

4.1 A map of the test area, the center of Hannover, Germany. . . . . 62

4.2 Landmarks with their reference regions. . . . . . . . . . . . . . . . 64

4.3 Representation of top levels of the hierarchical structure of land-marks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.4 Spatial distribution of Level 2 landmarks. . . . . . . . . . . . . . . 65

4.5 Graph representations of a grid network of named streets. . . . . . 70

4.6 Graph representations of named streets grid with a diagonal street—shortcut. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.7 Primal graph representation of a star-shaped street network. . . . 71

4.8 Dual graph representations of streets in a star-shaped street network. 72

4.9 Alternative streets with equal betweenness related to their suburbcontext. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

xix

LIST OF FIGURES

4.10 The graph of the connections of the level 2 districts by named streets. 76

4.11 Distribution of experiential rank values of the streets of Hannover. 77

4.12 33 most prominent streets of central Hannover. . . . . . . . . . . 79

4.13 Schema of relations between heterogeneous types of elements ofthe city in integrated hierarchies. . . . . . . . . . . . . . . . . . . 81

5.1 Schematic representation of the hierarchical partition of space. . . 91

5.2 The hierarchical representation of reference selection in CD. . . . 91

5.3 Process of selection of references for the destination descriptions. . 94

5.4 Selection of landmark references for the destination descriptions . 99

6.1 Route between the Universitat Hannover and the Staatstheater-Oper. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6.2 Turn-based route directions for the route between the UniversitatHannover and the Staatstheater-Oper by ©2007, Google Maps. . 114

6.3 Map of the route between the Universitat Hannover and the Staatstheater-Oper (©2007, Google Maps). . . . . . . . . . . . . . . . . . . . . 114

6.4 References selected for the destination description for the routefrom Universitat Hannover to the Staatstheater-Oper. . . . . . . . 118

6.5 References selected for the destination description for the routefrom Allianz-Hochhaus to the Katasteramt. . . . . . . . . . . . . 120

7.1 Destination descriptions transiting into turn-based route directionsin the proximity of the destination. . . . . . . . . . . . . . . . . . 134

xx

Chapter 1

Introduction

1.1 Motivation

1.1.1 Example

Current Web based map services, in-car navigation services and mobile naviga-

tion services are providing route directions to a variety of users in many contexts,

and for different purposes. Apart from traditional turn-based directions, with

references to street names and distance information, recent developments enable

the inclusion of point-of-interest (POI) and landmark information. Distance in-

formation is being replaced by more convenient estimates of travel times, along

with visualizations of the facades facing the street the wayfinder is navigating

along. The information provided is sufficient to identify a specific route, and

the significant actions that have to be taken along to route to follow it. The

information provided is thus sufficient even for a novice in the city.

A considerable proportion of the users of such navigation services consist of

people with at least coarse a-priori knowledge of the city in which they want to

navigate. Consider the following example of Stephanie, an inhabitant of Han-

nover, Germany, returning home from a business trip, and the directions she

provides to the taxi driver at the Airport:

Stephanie:“To Luisenstrasse, please.”

Taxi driver:“??”

Stephanie:“It is in the center, next to the Staatstheater, off Rathenaustrasse.”

Taxi driver:“Very well.”

. . .

Taxi driver:“Here is the Staatstheater, where should I go now?”

Stephanie:“It is that laneway after the theater. The house at the end, thank

you.”

1

CHAPTER 1. INTRODUCTION

Stephanie’s directions provide the taxi driver with a sketchy description of

the destination, without specifying the route to be take. Stephanie relies on the

taxi driver’s existing spatial knowledge, and only refers to prominent elements

of the city in the proximity of the destination. This form of communication of

spatial information is henceforth referred to as destination descriptions.

The contrast between destination descriptions and turn-based route direc-

tions can be illustrated by comparison with the directions provided to the taxi

driver by the navigation service of Google Maps (http://maps.google.com).

The turn-based route directions for the route from the Hannover Airport to

Luisenstraße are shown on Figure 1.1. Although other Web-based navigation

services may come with a different route, the characteristics of the route direc-

tions provided remain the same. The route is described based on the actions the

wayfinder has to take to reach the destination. This example highlights the signif-

icant differences between the hierarchical route directions composed by humans,

and the detailed, turn-by-turn route directions provided by current navigation

services.

1.1.2 The Familiar Structure of the Environment and Route

Directions

Consider the example of Stephanie and the taxi driver. A close look at the

references used reveals a spectrum of different types of referents: regions, such as

the city, notable buildings (the Staatstheater), and linear referents along which the

wayfinder can navigate (Rathenaustrasse,the laneway). As we can see, references

to several types of elements of the city (Lynch, 1960) may occur in one set of

human-generated route directions. This is in contrast to the traditional approach

of current navigation services, that typically only include references to paths.

Furthermore, we can notice that Stephanie does not start her route directions

at her current position (at the airport), but instead refers to a distant and vaguely

delimited region distant from the airport—the center. She not only uses an

unofficial label for the district, but she also assumes that the taxi driver will be

able to identify it and knows how to get there. Her route directions proceed

from a reference to the larger neighborhood of the destination in steps closer and

closer to the destination, with each consecutive referent having lesser and lesser

probability of being known by the taxi driver. Stephanie does not enforce an

exact route to the taxi driver.

2

M. TOMKO 1.1. MOTIVATION

(a) Overview of the route and turn-based route directions.

(b) Detail of the route destination.

Figure 1.1: Turn-based route directions from Hannover Airport to Luisenstrasse(©2007, Google Maps).

3


1.1.3 Inferential Communication

Let us consider the above example in more detail. Stephanie and the taxi driver

have never met before. Thus, Stephanie does not have any information about

the taxi driver’s spatial knowledge, nor does the taxi-driver have any information

about Stephanie’s spatial knowledge. If the classical communication theory which

assumes that all the information needed for understanding a message is contained

within the message (Shannon and Weaver, 1949), held true, the route directions

provided by Stephanie would not be sufficient to guide the taxi-driver to the

destination, as they do not contain the full information about the actions that

the wayfinder must take to reach the destination. The taxi-driver, however,

understands the meaning of the directions provided by Stephanie and gets her

home.

Stephanie makes several assumptions about the taxi driver, as she enters

the cab. These assumptions are made based on observations and common sense

knowledge of the situation. Stephanie may assume that the taxi driver has some

spatial knowledge of Hannover, assumes that the taxi driver will try to be coop-

erative by interpreting the directions provided by Stephanie, and that the cab in

which she is sitting will be used to get her home.

Many more, and finer grained assumptions can be made by Stephanie. Based

on these assumptions derived by inference from the situation at hand, Stephanie

provides route directions to the taxi driver. The assumptions made by Stephanie

are necessary to correctly interpret the situation in which the communication

occurs and provide information to the taxi driver in a way that will be understood.

People experience deficiencies of the classical communication model every

day. Linguists have addressed these deficiencies by developing theories of prag-

matic communication. This thesis explores situations where the referents included

in route directions exchanged between agents are assumed to be part of the com-

mon knowledge. The approach is grounded in the relevance theory of inferential

communication (Sperber and Wilson, 1986). Relevance theory is built on a gen-

eral view of human cognition with the assumption that human cognitive processes

tend to maximize the efficiency of any action. The relevance of possible referents

is evaluated, and the referent which satisfies the requirements for relevance in the

given situation is selected. This thesis aims to present a formal model of this

selection process.

4

M. TOMKO 1.1. MOTIVATION

1.1.4 Current Approaches to Route Directions’ Personal-

ization

Current research in navigation services focuses mostly on three broad areas in

which methods of personalization are studied: route planning, concerned with

planning better routes for specific users or groups of users (Caduff and Timpf,

2005; Duckham and Kulik, 2003; Haque et al., to appear 2007) or better user

interfaces, possibly with multi-modal interaction options, such as natural (spoken)

language (Dale et al., 2005) and content adaptation.

Google Maps, Yahoo Maps and other similar Web navigation services gener-

ate turn-based route directions, providing complete information of the actions the

wayfinder has to perform to follow the routes retrieved by the respective services.

The service MapQuest implements a route generalization algorithm LineDraw by

Agrawala and Stolte (2001), presenting the route with emphasized regions of high

complexity. Recent research of Klippel et al. (2003) and Richter (2007a) focused

on chunking of route directions based on the structural properties of the route

described, in order to decrease the number of information items in the result-

ing directions. The adaptation of the level of detail of the directions provided

is, however, determined purely by the route structure and does not consider the

possibility of the user’s partial familiarity with the environment.

In general, the focus of current services, as well that of major research ef-

forts is on wayfinders without previous experience with the environment. Locals,

however, may find turn-based directions excessive and patronizing. The level of

detail of the information provided by current navigation services is not appropri-

ate for users with previous spatial knowledge of the environment through which

the route leads. In the example of Stephanie and the taxi-driver, the information

contained in the last two instructions of the Google Maps route planning engine

is equivalent to the directions provided by Stephanie. The rest of the information

provided is excessive for the taxi driver, leading to a high cognitive workload

with no advantage or improvement of the wayfinding success. While in the case

of a taxi-driver this may not be critical, a driver of an ambulance would greatly

benefit from a system reducing the information overload. In today’s dense and

fast traffic, an ambulance driver driving fast to a casualty needs to concentrate

on driving, rather than on following directions. Furthermore, destination descrip-

tions similar to those frequently provided by humans are not prescriptive, even

allowing for route alterations in case of a changed traffic situation or a traffic jam.

This thesis focuses on the formalization of the process of selection of refer-

ences to environmental features, which are relevant in destination descriptions for

5


wayfinders with at least partial spatial knowledge of the environment. The bene-

factors of the research presented in this thesis are developers of navigation systems

for wayfinders frequently travelling to unfamiliar destinations in partially-known

urban environments, such as the police, emergency management and tourism ser-

vices, but also locals—everyday users of Web based navigation portals, location-

based services and in-car navigation services.

1.2 Scope and Objectives

The goal of this research is to analyze how the structure of an urban environment

is reflected in the way humans describe route destinations, and to formalize these

principles in a computational model that would enable automated creation of

destination descriptions. The research question explored in this thesis is:

What is the relation between the city structure and the route directions of

familiar wayfinders?

The research presented in this thesis builds on the hypothesis that it is possi-

ble to construct a generic model of destination descriptions based on the knowl-

edge of the inherent functional city structure.

The answer to the research question leads to a formal model for the genera-

tion of destination descriptions for persons with previous spatial knowledge of the

environment. The objective is to identify the referents that are likely to be used

in the given context by humans. The model is based on the basic assumption of

pragmatic behaviour that rational beings try to limit their cognitive effort and

energy expenses while still being able to reach a goal. People try to minimize their

effort by being as efficient as possible while carrying out a task. In the context

of communication, the speaker is trying to transfer the information to the hearer

through a message while minimizing the effort required to do so. The recipient

of the information interprets the meaning of the message, also trying to minimize

her or his effort. In the situation of communication of destination descriptions

the communication partners mutually assume the possession of a-priori spatial

knowledge of the environment. Consequently, they refer to it in the destination

descriptions. The effort of both the speaker and the hearer is thus decreased.

This thesis focuses on the identification of relevant references in a given con-

text. The context is defined by a restricted set of characteristics: the spatial

context characterized by the start and destination of the route; and the commu-

nicators, two mutual strangers exchanging destination descriptions (and thus can

not refer to places previously experienced together). The certainty of the correct

6

M. TOMKO 1.3. APPROACH

identification of the referent by the hearer can thus not be granted through mu-

tual experience. It is not the goal of this thesis, however, to emulate the exact

behavior of specific humans, nor is the goal a cognitive study of the resulting

route directions.

1.3 Approach

The approach taken in this research starts from the empirical characterization

of human generated destination descriptions, such as Stephanie’s, allowing the

research hypothesis to be formulated. The comparison of the characteristics of

destination descriptions with the structure of space in which they are provided,

and with the human spatial mental representations identified in previous works,

the common properties of destination descriptions are identified. Once the charac-

teristics of route directions and the spatial, communicative and cognitive context

in which they are created are determined, the processes of the construction of

destination descriptions can be studied. The identification of references for des-

tination descriptions will be based on the characteristics of the spatial structure

in which they occur.

The approach applied builds on the interdependence of state and process de-

scriptions of complex systems (Simon, 1962), a common problem-solving method

in cognitive science and artificial intelligence research. Starting with a state

description of destination descriptions, the characteristics of the construction

process of destination descriptions is inferred. The process description is then

formalized, computationally implemented and applied on a test dataset. The

characteristics of the outcomes of the model are compared with the characteris-

tics observed empirically. If they are equivalent, the model is considered valid.

From the observation of similarities in route directions communicated by

people familiar with the environment, it can be concluded that common principles

are followed in the process of selection of references. The general characteristics

of such route directions are collected and form the basis of the hypothesis.

The empirical characteristics gathered are further interpreted with regard to

communication theory, psychology and cognitive science. The principles inferred

from this body of knowledge are then formalized in a general model of selection

of references in destination descriptions. The model consists of a set of rules

determining the selection process.

The constraints imposed on the model are limited to the hierarchical spatial

data model of the urban environment, without individual considerations of a

7


more detailed wayfinder’s profile. Thus, the model should provide plausible route

directions for wayfinders satisfying the condition of at least partial a-priori spatial

knowledge of a given urban environment.

The conceptual model is consecutively implemented in a computationally

executable model. The inputs are limited to the hierarchically structured dataset

of the environment and the route to be described. No other inputs are allowed

during the execution of the model implemented. The output of the model is

the the set of references constituting the destination description. This output is

verified against the specification of the previously gathered characteristics.

1.4 Expected Outcomes

The main objective of this research is to understand how the structure of an

urban environment is reflected in the way humans provide directions to famil-

iar wayfinders. The formalization of these principles in a computational model

is the main outcome of this research. A set of heterogeneous environmental ele-

ments of the city structure is considered in a concise model of destination descrip-

tions in route directions, together with a theoretical model of the communication

processes underlying the selection of the referents. The computational model

enables an automated creation of route directions with varying level of detail in

urban environments where hierarchically structured data can be made available.

The model thus allows the implementation of navigation systems filling the gap

between human generated route directions and the turn-based route directions of

current Web-based or on-line navigation systems.

1.5 Thesis Structure

This thesis is structured as follows: the next Chapter provides an overview of

relevant concepts and existing literature in the fields of spatial cognition, route

planning and wayfinding (Section 2.1) and communication theory, especially from

the perspective of inferential communication and the use of common knowledge

(Section 2.2). This section concludes with a discussion of modeling and formal-

ization of complex systems (Section 2.3).

Chapter 3 introduces the concept of destination descriptions, based on a

simple example scenario. This example is used to introduce the definition and

characteristics of destination descriptions, and the role of common spatial knowl-

edge in destination descriptions of locals.

8

M. TOMKO 1.5. THESIS STRUCTURE

In the following chapter (Chapter 4), experiential structures of spatial knowl-

edge are introduced as a necessary input for an implementation and testing of a

formal model of selection of references for destination descriptions. Furthermore,

the concepts presented are applied to construct an experiential, integrated and

hierarchical dataset of Hannover.

In Chapter 5, the pre-requisites for the model of selection of references for

destination descriptions are stated, and the model is formalized. The model is

presented as a set of rules allowing to select relevant references for destination

descriptions. These rules are serialized in a set of algorithms describing their

execution.

The model presented is then implemented and tested on a test dataset of

central Hannover. The computational implementation and the results of the

tests are discussed in Chapter 6.

The thesis concludes with Chapter 7, where the main contributions of the

thesis are discussed, and the outlook to future research is drafted.

The thesis is completed with several appendices providing more detail for

the interested reader. Appendix A and Appendix B contain details about the

test dataset, while Appendix C contains the code of the Haskell implementation

of the model. Finally, Appendix D contains the inputs and outputs of example

test cases of the model.

9


10

Chapter 2

Background

This chapter reviews previous work in the fields of spatial cognition, commu-

nication and computer science related to the acquisition and communication of

spatial knowledge.

People store their spatial knowledge in mental representations of space. This

environmental knowledge is acquired through interaction with the environment

and facilitated through perception. Mental representations of spatial knowledge

and their organization are discussed in Section 2.1.

Communication in and about space, such as direction giving, represents an

important use of people’s spatial mental representations. People familiar with

an environment share common spatial knowledge due to similar experience of

their environment. This knowledge is then used in place and route descriptions

they exchange. Concepts from the pragmatic theory of communication are intro-

duced in Section 2.2, and point to the significance common knowledge plays in

communication.

Section 2.3 provides an introduction to modelling and formalization and is an

introduction to the methods used in operationalization of the theoretical model

of destination descriptions presented in this thesis.

2.1 Experiencing Space

2.1.1 Mental Representations of Space

People learn the layout of their environment, be it natural or built, through con-

tinuous interaction. They perceive the environment through senses, learn the

layout of the environment and store the knowledge acquired in mental represen-

tations (Fig. 2.1).

11

CHAPTER 2. BACKGROUND

Figure 2.1: Acquisition of spatial knowledge. A person perceives the environment,learns its layout, and stores this knowledge in a mental representation (adaptedfrom Frank (2000) and Frank et al. (2001)).

With repetitive interaction, people get increasingly familiar with the layout

of the environment and develop spatial knowledge stored in spatial mental repre-

sentations (Siegel and White, 1975). With continuous interaction, the structure

of this spatial knowledge evolves through stages with different characteristics.

At first, landmark knowledge is acquired, characteristic by discrete knowledge

of salient spatial features. The spatial relations between these features may

not yet be established. Landmark knowledge is integrated into more complex

structures—sequences, also called routes. Hence, at this stage of spatial knowl-

edge evolution, route knowledge is formed. At this stage, landmarks are recalled

in the order as experienced along a learned route, and their complex spatial rela-

tionships may not be evident. As people become familiar with the environment,

survey knowledge is formed, enabling them to locate and infer directions and

distances (metric properties) between the individual spatial features.

Tversky (1993) proposed to distinguish spatial mental representations (re-

ferred to as mental maps) into further sub-categories of cognitive collages and

spatial mental models. Cognitive collages are patchy in nature, contain only

partial information on the environment, and are heavily distorted. In contrast,

spatial mental models allow integration of spatial knowledge from distant regions

as well as perspective taking and inference of directions between landmarks. Spa-

tial mental models may still be metrically inaccurate. Spatial mental models were

suggested to represent the spatial mental representations of familiar environment.

The speed with which people learn the environment and transit between

these stages of spatial mental representations is highly individual, and is largely

depending on their spatial abilities (Allen, 1999) and the frequency of interac-

tion with the environment. Continuous interaction with the environment allows

to proceed from landmark to route to survey knowledge. It has been recently

hinted, however, that these types of knowledge may be acquired simultaneously

(Ishikawa and Montello, 2006), depending on the individual’s spatial abilities. For

instance, some individuals may acquire basic metric characteristics of space after

relatively little interaction. Thus, continuing interactions allows the accuracy and

12

M. TOMKO 2.1. EXPERIENCING SPACE

completeness of the acquired spatial knowledge to improve, although individual

differences in spatial abilities effectively impose a limit to such improvements.

Thus, individuals with innate good spatial abilities or good spatial training (e.g.

geographers (Golledge et al., 1995), taxi drivers) may quickly form accurate and

relatively complete spatial mental models of the environment. This progression

happens, however, not in months or years as previously suggested, but within a

few trips through the environment in question (Ishikawa and Montello, 2006).

The mental representation acquired through direct experience of the environ-

ment is further supported by secondary, indirect spatial learning. Inhabitants of

a city enhance their spatial knowledge from sources such as maps, news articles,

advertisements and Web resources. These spatial narratives (Levine and Klin,

2001; Weissensteiner and Winter, 2004) and geo-referenced descriptions (Winter

and Tomko, 2006) add to the spatial mental representations of people residing in

a specific city during longer periods of time.

2.1.2 Conceptualization of Space

Following the representational theory of mind, the basic elements of mental rep-

resentations are called concepts (Margolis and Laurence, 2006). Concepts are the

result of a cognitive process of categorization of the knowledge acquired while

perceiving the world (Rosch, 1978). The goal of the cognitive process is to create

a simplified, abstract model of the knowledge acquired, in order to reduce its

complexity and thus the mental effort required to store it. Furthermore, con-

ceptualizations allow generalizations and abstract reasoning about the domain of

knowledge processed.

Mental conceptualizations of space and of spatial phenomena have been the

subject of intensive research (e.g., Downs and Stea, 1977; Freksa and Barkowsky,

1996; Hirtle, 2003; Klippel, 2003b; Klippel et al., 2004, 2003; Klippel and Winter,

2005; Lynch, 1960; Montello et al., 2003; Richter and Klippel, 2005). An essential

contribution to the conceptualization of space is the work of the urbanist Kevin

Lynch, studying the phenomenon of imageability of urban environments. In his

experiment, Lynch studied the composition of sketches provided by the inhabi-

tants of American cities (Lynch, 1960). Sketch maps represent common devices

people use to communicate spatial knowledge (Agrawala and Stolte, 2001), and

as such present a convenient and familiar way of externalizing spatial mental rep-

resentations in cognitive research (e.g., Kim and Penn, 2004). The analysis of

the sketches collected by Lynch revealed five basic structural elements of the city

form: paths, nodes, landmarks, districts and edges. These elements present the

basic concepts in the (static) spatial mental representations, i.e. the image of the

13


city people retain after experiencing it.

Briefly, Lynch defines the elements of the city form as follows:

• Path: a path is a one-dimensional physical entity of the environment along

which observers move. The city is experienced through movement along

paths. From the functional point of view, any network facilitating transport,

such as the street network, water canal network or railroad network can

represent a network of paths.

• Node: nodes are points, strategic foci of activity in which the observer

can enter. Typically, nodes are convergences of paths (junctions), where

concentrations of characteristic activities occur. As noted by Lynch, some

nodes are the focus and epitome of a district, of which they stand as a

symbol.

• Landmark: Landmarks are point-like spatial features, serving as spatial

references. The observer cannot enter within a landmark, it can only be

experienced—observed—from the exterior. According to Lynch’s definition,

landmarks stand out from their environment, thus being salient. Land-

marks present convenient clues in the spatial structure of the imagined

environment. Lynch distinguishes between distant landmarks, representing

inaccessible locations, and local landmarks, visible only in a restricted space

and from a certain viewpoint.

• District: defined by Lynch as medium-to-large sections of the city with a

two dimensional extent. Districts share some common characteristics and

have a distinct inside and outside. Districts and paths are singled out as

the most dominant and distinctive elements of the city form.

• Edge: In Lynch’s terminology, an edge is a linear element not considered as

path. Lynch states that they are usually the barriers between two kinds of

areas, or districts. They are impenetrable to cross-movement. The effect of

edges on the perception of the hierarchical structure in the urban environ-

ment is closely related to the boundaries between districts. The term edge

can be used interchangeably with barrier.

Lynch’s urbanistic analysis of elementary spatial concepts was based on a

medium scale view on the city and did not consider a more detailed structure

within the urban fabric of the urban environment. However, human hierarchical

reasoning is flexible in its ability to change the granularity levels of mental repre-

sentation across multiple granularities. Humans identify patterns and structures

within limited, smaller parts of the city, as well as in its larger region. This

14


encourages the theory that the principles leading to the formation of a coarse

representation of the city also contribute to the representation of its parts. This

hypothesis of scalability (Carrera, 1997) explores the application of Lynch’s basic

urban morphemes to the analysis of a single square. Other attempts to re-define

spatial concepts constructing the structure of an urban space can be found in

the work of Singh (1997), specifically focusing on nodes defined as areas where

districts of multiple land use types meet.

It is necessary to note that the term landmark may also be used for any of

the type of elements of the city, if the element in question holds the quality of

landmarkness, i.e. is prominent due to salience of its characteristics in a given

environment (Raubal and Winter, 2002). The term landmark will be further used

according to Lynch, to refer to point-like features of high salience. It is implied

that all references to spatial features found in route direction have the property

of salience or prominence, standing out from the environment.

Furthermore, it is important to note the difference between physical features

of the world having the function of a path (i.e. locomotion occurs along them)

(Lynch, 1960; Montello, 2005), and the notion of paths used to design the results

of a path planning process (mental or computational). The terms route planning

and route will be further used in such cases, following the definition of Montello

(2005). Montello (2005) defines a route as a linear pattern of movement of a

wayfinder, as opposed to paths, a linear physical feature along which movement

occurs (also see Section 4.3). Although in urban environments the movement

along a route typically occurs on paths, this is certainly not a general limitation.

In less structured built environments (parks, squares) and natural environments,

wayfinders are not bound to paths (e.g. when crossing a meadow).

References to all five types of elements identified by Lynch may occur in

route directions communicated by people. The definitions of each of the type

of elements provided by Lynch are, however, vague. Refined definitions of the

elements of the city are therefore used for operationalization in the model of

destination descriptions (Section 3).

2.1.3 Hierarchical Structure of Spatial Mental Represen-

tations

Individual mental representations are distorted due to spatial abilities, individ-

ual experience of space and the cognitive processes leading to their construction.

The nature of these distortions provides valuable hints regarding the internal

organization of spatial knowledge in spatial mental representations. Individual

15


spatial behavior, cognitive capabilities and responses to specific properties of the

environment are the causes of the distortions in these representations (Stevens

and Coupe, 1978). Couclelis et al. (1987) suggest a hierarchical relation between

spatial cues and their areas of influence, where important spatial cues—anchor

points—were found to be foci of so called tectonic plates, regions in which the

cues tended to be strongly associated together. Distortions in representations of

the location of the anchor point were transmitted on the cues associated. Fur-

ther research confirms this hierarchical organization of spatial knowledge (Hirtle,

2003; Hirtle and Jonides, 1985; Taylor and Tversky, 1992). The hierarchical or-

ganization of mental representations is reflected in spatial reasoning tasks, where

dependence between the categorization of a specific spatial entity in the hier-

archy and its use in the spatial task was demonstrated (Plumert et al., 1995a;

Timpf and Kuhn, 2003; Wiener and Mallot, 2003). A graph based approach of

the development of such hierarchical spatial knowledge through wayfinding was

proposed by Chown et al. (1995), and implemented in a system called PLAN.

Mental spatial hierarchies are not likely to follow discrete, well defined hier-

archical levels as they are used in e.g. hierarchical data structures such as quad-

trees. While in computing hierarchical data structures are frequently adopted for

efficient retrieval of exact information, hierarchies in mental conceptualizations

emerge to lower the cognitive effort of storing and retrieval of the information.

The retrieved information may often be approximate, as far as it is sufficient

to support the task of the agent. The formation of chunks of information is a

means of lowering the cognitive effort and adapting the complexity and quan-

tity of information that has to be remembered (Cowan, 2001; Miller, 1956). As

further mentioned by Taylor and Tversky (1992), grouping, hierarchical organi-

zation and coherent linking contribute to comprehensibility. Thus, hierarchical

grouping of spatial information in mental representations is a means to effectively

cope with the complexity of the environment (Klippel et al., 2003; Tomko and

Winter, 2006a).

The necessity to narrow down the processing space (or the space of interest)

as a section of the surrounding environment is reflected in vague spatial concepts

such as far and near, derived from the physical and perceptual accessibility of the

places referred to. Montello (1993) proposes a hierarchically ordered framework

of psychological spaces, consisting of four major classes: figural, vista, environ-

mental and geographical. These classes are ordered by perceptual and physical

accessibility by the embodied agent. Figural space is projectively smaller than

the size of human body, while vista space is equivalent or larger, although still

perceivable from a single viewpoint. Environmental space is larger than human

body and surrounds it completely. It is too large to be perceived at once, and

16


from a single viewpoint. Montello (1993) suggests that environmental space needs

to be learned by integrating the information acquired over significant periods of

time. Such spaces are then equivalent to cities and city neighborhoods. Finally,

spaces of geographical scale are projectively much larger then the human body

and cannot be learned by locomotion. Such spaces must be learned through

symbolic representations, such as maps. This thesis focuses on destination de-

scriptions in urban spaces of environmental scale. The performance of the model

of destination descriptions proposed in geographical spaces is possible, although

not tested.

2.1.4 Route Planning and Wayfinding

Once formed, spatial knowledge is used in day to day interaction with the envi-

ronment, such as route planning and wayfinding. Route planning is the mental

activity of planning the way through an environment. As noted by Couclelis

(1996), two stages of route planning can be distinguished. The first, coarse plan-

ning stage consists of the retrieval of a detached mental view of the area, resem-

bling a map-like view. Major structural elements of the environment as well as

channels between the start and the destination are retrieved from memory. This

stage is most likely to occur among subjects with high spatial abilities or good

familiarity with the environment. In hierarchical, regionalized environments, the

influence of important higher-order regions on route planning has been confirmed

(Wiener and Mallot, 2003).

In the next, fine-level route planning stage, a mental representation of the

route is developed, involving detailed, first person’s imagined travel through the

environment, including embodied actions at decision points. This representation

is then used during wayfinding (and compared to the environment perceived by

the agent during locomotion), or in the communication of route directions (Frank

et al., 2001).

In case of unfamiliar environments, spatial knowledge needed for wayfinding

has to be acquired prior to wayfinding. Other sources, such as maps and route di-

rections are used to form a mental representation of the environment (Figure 2.2).

These source are a form of communication between agents with a-priori spatial

knowledge of the environment (e.g., local experts, but also automatic route ser-

vices accessing spatial databases) and agents receiving this spatial knowledge

through communication without a-priori spatial knowledge (e.g., tourists). The

expert agent retrieves a route in a route planning process based on their own

expert judgment, or based on criteria specified by the wayfinder. A formal model

linking the stages of route planning with route directions and the navigation as

17


such was proposed by Timpf et al. (1992).

Figure 2.2: Wayfinding in unfamiliar environment. A hearer’s spatial mental rep-resentation (Map 2) is acquired by communication from a knowledgeable agent,the speaker.

In situations where a speaker is providing route directions in the form of

destination descriptions to a wayfinder with a-priori spatial knowledge of the

environment, the planning process is performed independently by the wayfinder,

combining the information contained in the destination description with his or her

own spatial knowledge (Section 2.2.6). The route resulting from the internal route

planning process of the speaker need not be identical to that of the wayfinder.

In recent years, the development of route planning services led to research

in, and specification of, route planning algorithms retrieving routes optimized to

satisfy characteristics more complex then length of the route, as is the case for

the well known Dijkstra and A∗ algorithms. Of special interest are algorithms

allowing the retrieval of low complexity routes (simplest paths) (Duckham and

Kulik, 2003; Mark, 1986) or those that can be described with least ambiguity

(Haque et al., to appear 2007).

With the increasing size of spatial datasets, the performance of algorithms

became an issue. Hierarchical route planning algorithms were proposed, allow-

ing the speed of calculation to be optimized by reducing the search space (Car,

1997; Jung and Pramanik, 2002). Such approaches are inspired by human hi-

erarchical spatial reasoning, where the overall planning of a route from fine to

coarse (the so-called skeleton) and back again to finer levels of detail was observed

(Kuipers, 2001). The route planning process seeks to find a route from the local

streets of low hierarchical level to hierarchically important roads representing the

skeleton of the road network. In the proximity of the destination, the inverse

process searches for local streets, allowing to reach the destination. While in

computational systems this approach gains importance with the increasing size

of the road networks, people employ hierarchical heuristics for route planning

in comparatively smaller regions, due to the cognitive complexity of retrieving a

18

M. TOMKO 2.2. COMMUNICATION

route.

2.2 Communication

This section introduces relevance theory (Sperber and Wilson, 1986), a funda-

mental communication theory in which the model of destination descriptions

presented in this thesis is grounded. After an introduction to communication

theory in general, and relevance theory in particular, the implications to spa-

tial communication are drawn, with the focus on place descriptions and route

directions.

2.2.1 Communication Models

Research in information technology, and certainly research in spatial information,

was predominantly built upon the classical, or code model of communication,

introduced by Shannon and Weaver (1949). The code model of communication

assumes a communicator—speaker—that encodes the information to be conveyed

to the receiver (hearer). The well known semiotic triangle relates a referent, its

conceptualization in a mental representation, and the generated symbol (Ogden

and Richards, 2001). This symbol is then used by the speaker in communication

to convey the information about the referent. The information encoded in a

message is transmitted through a communication channel and decoded by the

hearer. Such a semiotic process is engaged in the encoding of the information

into the message by the speaker and in the decoding of the symbols from the

message by the hearer. In this thesis, the term reference is used instead of the

term symbol. The term referent is used for spatial elements that are referred to.

The central assumption is that the totality of information received by the

hearer was encoded by the speaker and transferred through the channel. Further-

more, the code model of communication provides for loss of information during

the transmission of the message along the channel by the influence of information

noise, or entropy.

As Worboys (2003) notes, the classical model of Shannon and Weaver does

not sufficiently account for several fundamental influences on communication of

spatial information, such as characteristics of the channel and context of the

hearer and the speaker. These deficiencies have been noted by linguists before,

especially when studying conversations and noting the linguistic context of utter-

ances. The deficiencies of the classical communication model have been addressed

in pragmatic theories of communication.

19


2.2.2 The Meaning of Utterances

Pragmatic theories of communication deal with utterances, which are defined as

specific intentional acts of speaking. Pragmatics analyse the meanings of utter-

ances, such as the meaning intended by the speaker, the meaning conveyed to the

hearer, the meaning understood by the hearer, as well as the circumstances of the

utterance. The fundamental difference between the classical theory and pragmat-

ics is in the realization that the explicitly encoded and transmitted content of a

communication is influenced by the context of the communication, the intentions

of the communicating parties, and other external and internal influences. The

content of a message conveyed may therefore be much richer then the one explic-

itly encoded in the message, for instance by implying additional meaning. Grice

(1957) defines the meaning of an utterance to be the cognitive effect produced by

the recognition of the intention of the utterance. Cognitive effect is defined by

Sperber and Wilson (1986) as the contextual implication caused by the stimulus

in a cognitive system. Thus, the cognitive effect is achieved in function of the

context (or cognitive environment) and the stimulus.

The realization that the meaning conveyed by an utterance may be funda-

mentally different to the meaning intended, or to the meaning conveyed, is one

of the important realizations of pragmatic theory relevant to this thesis. The

meaning implied by utterances was studied in the seminal works of Grice (1957,

1975, 1989). Grice formulated his cooperative principle:

Make your conversational contribution such as is required, at the

stage at which it occurs, by the accepted purpose or direction of the

talk exchange in which you are engaged (Grice, 1989).

Grice’s cooperative principle is operationalized by the speaker in a conversa-

tion by following four conversational maxims:

• quantity (“Make your contribution as informative as required by the pur-

pose of the exchange; do not make your contribution more informative than

is required.”);

• quality (“Do not say what you believe to be false; do not say that for which

you lack adequate evidence.”);

• relevance (“Be relevant.”)

• manner (“Avoid unnecessary prolixity; avoid ambiguity; be brief; be or-

derly.”).

20


Grice assumes that the hearer rationally applies the maxims as rules of thumb

in understanding the implied meaning, carried by an utterance. The intended

meaning may not be conveyed by the speaker if the maxims are not respected.

The utterance may convey a changed meaning to the hearer, resulting in an im-

plicature. The recognition of the intentions of the speaker, oriented toward the

hearer in an overt manner is yet another contribution of Grice to pragmatics.

While largely discussed, Grice’s theories laid the foundations of modern prag-

matics.

2.2.3 Relevance Theory

The decoding of the meaning of an utterance is not satisfactorily explained by

Grice’s contributions. The understanding of the implied content requires another

reasoning step, that is the inference of the speaker’s intentions by the hearer.

The inferential communication model is at the core of the relevance theory of

Sperber and Wilson (1986), grounded on a model of human cognition. It uses

Grice’s concept of relevance in a much extended manner, and explains inferential

communication purely in terms of this concept.

Relevance theory emphasizes the importance of the cognitive environment

to the comprehension of an utterance. It defines a cognitive environment as

a set of assumptions available to a cognitive agent. In this environment, the

act of communication presents the act of construction of a verbal or non-verbal

stimulus, a phenomenon meant to achieve cognitive effects. This stimulus triggers

expectations of relevance by the act of changing the cognitive environment itself.

This conforms with Grice’s observation that the very fact of communicating

triggers, among the audience, expectations that it then exploits (Sperber and

Wilson, 1986, 37). It is, however, only valid in cases of ostensive communication,

i.e. when the act of communication is made explicitly manifest to the hearer.

The act of ostension itself guarantees relevance to the hearer. Relevance theory

is concerned with ostensive inferential communication. Relevance theory builds

upon two basic observations about relevance in cognition and communication: the

cognitive principle of relevance (Definition 1) and the communicative principle of

relevance (Definition 2), defined as follows:

Definition 1 Cognitive principle of relevance: Human cognition tends to be

geared toward maximisation of relevance.

and

21


Definition 2 Communicative principle of relevance: Every act of overt commu-

nication conveys a presumption of its own optimal relevance.

A stimulus is relevant if it connects with available contextual assumptions to

provide a positive cognitive effect. Of course, many stimuli of varying relevance

may be perceived by an individual at any time. Thus, building on the two pre-

vious principles, cognitive agents will pick the most relevant stimulus in a given

communication situation. It implies that the hearer is the one that interprets a

reference in a communication in the most relevant manner to him or herself. The

speaker, on the other hand, makes sure that the stimulus is perceived as relevant,

through content or form.

Based on these starting points, Sperber and Wilson (1986) formulate the

presumption of optimal relevance:

• The utterance is relevant enough to be worth processing

• It is the most relevant one compatible with the communicator’s abilities

and preferences

From there, the hearer can infer the meaning of the utterance. Again, several

meanings may be inferred (Sperber and Wilson, 2004, p.3):

“Intuitively, relevance is not just an all-or-none matter but a mat-

ter of degree. There is no shortage of potential inputs which might

have at least some relevance for us, but we cannot attend to them all.

Relevance theory claims that what makes an input worth picking out

from the mass of competing stimuli is not just that it is relevant, but

that it is more relevant than any alternative input available to us at

that time. Intuitively, other things being equal, the more worthwhile

conclusions achieved by processing an input, the more relevant it will

be.”

Sperber and Wilson use the expression degrees of relevance, implying that

certain stimuli may be more or less relevant then others, depending on the ration

of the processing effort and contextual implications they afford. The maximiza-

tion of relevance leads through a path of minimal effort, as living organisms tend

to minimize the expenditure of energy in every situation. As Sperber and Wil-

son (1986) state, human cognitive processes are geared toward maximizing the

cognitive effect of a stimulus, while minimizing the cognitive effort necessary to

process it:

22


Definition 3 The principle of relevance:

• Everything else being equal, the greater the cognitive effect achieved by the

processing of a given piece of information, the greater its relevance for the

individual who processes it.

• Everything else being equal, the greater the effort involved in the processing

of a given piece of information, the lesser its relevance for the individual

who processes it.

Some of the meanings that can be inferred from the utterances require less

processing effort, and are thus more easily derived (Noveck and Sperber, 2006)

in a given context. This meaning represents the interpretation of the utterance

that is most naturally derived. Relevance theory assumes that hearers follow

heuristics of comprehension build on the principles stated above. Thus, they

follow the paths of minimal effort to interpret an utterance, infer its meaning in

a given cognitive environment, and stop as soon the meaning found satisfies the

expectation of relevance.

Of course, a different set of stimuli in the environment of the speaker may

lead to a different meaning inferred by the hearer. Also, the speaker does not have

to be optimally relevant in an utterance, as a better stimulus may be found in a

given situation. It only has to appear to the hearer that the utterance is optimally

relevant. The hearer may not know that the speaker may be able to construct a

stimulus of higher relevance; the hearer just believes that the stimulus received

is optimally relevant. Such deception is a frequent tool of lawyers in legal text,

where a statement may be deliberately worded in a manner allowing multiple

interpretations. And indeed, even in less critical situations, it has been shown

that speakers are not always trying to be cooperative to the full possible extent.

Even then, however, understanding can be achieved (Davies, 1995).

2.2.4 Knowledge, Context and Communication

In human communication, people share understanding of the world surround-

ing them. This environment is perceived and cognitively processed, it is thus a

cognitive environment, also known as context. This understanding significantly

influences the interpretation of information received in communication. An im-

portant part of the cognitive environment consists of knowledge previously ac-

quired, be it in previous utterances (i.e. linguistic context), or by interaction with

the environment. Sperber and Wilson (1982) defines context as follows:

23


Definition 4 Context . . . is the background information that can be brought to

bear on comprehension.

Sperber and Wilson, however, further note that what context consists of is difficult

to define. A common approach to context-adaptive information system therefore

usually consists of detailed definitions of relevant parameters of context.

In inferential communication, the speaker and the hearer may assume the

possession of tacit knowledge by the communication partner as an important

part of the communication partner’s cognitive environment. Tacit knowledge is

the part of one’s knowledge that has not been made explicit in the information

exchange. Tacit knowledge was identified as an important influence on commu-

nication in situations of co-ordination and collaboration. Researchers suggested

a theory of mutual knowledge to explain the use of tacit knowledge by commu-

nicators (e.g., Clark and Carlson, 1982). There, mutual knowledge (also mutual

belief) is defined as the kind of knowledge that is the product of an infinite series

of reciprocal expectations. Let us imagine two agents, A and B, and a mutual

belief of a phenomenon p:

Definition 5 Mutual knowledge

(1) A knows that p.

(1’) B knows that p.

(2) A knows that B knows that p.

(2’) B knows that A knows that p.

(3) A knows that B knows that A knows that p.

(3’) B knows that A knows that B knows that p.

. . .

Clark and Marshall (1981, cited by Sperber and Wilson (1982)) mention

physical co-presence, linguistic co-presence and community membership as factors

allowing the inference of mutual knowledge. The already complex establishment

of mutual knowledge becomes exponentially difficult with an increasing amount of

communicators (Clark and Carlson, 1982). Relevance theory therefore refuses the

concept of mutual knowledge on basis of cognitive effort, as it is implausible that

such a complex and infinite reasoning process would be necessary in frequent and

trivial communication tasks. Meaning inference is offered instead as a solution

(Sperber and Wilson, 1982).

24


Some sort of existing knowledge, however, is undeniably present and referred

to in many coordination tasks, such as reference to objects, familiar place descrip-

tions and route directions. The term common knowledge, also known as shared

knowledge or common belief, is therefore used for knowledge that a communicator

assumes is known to all partners in communication. Note that the requirement

of mutuality is thus relaxed.

The factors identified by Clark and Marshall (1981) (physical co-presence,

linguistic co-presence and community membership) are necessary for the inference

of common knowledge of the communication partners in a given communication

context. These factors thus assist with the inferential identification of references

used in communication. A detailed taxonomy of co-presence is offered by Zhao

(2003), based on two factors: co-presence as mode of being with others, and

co-presence as sense of being with others. In this thesis, co-presence will be

operationalized as an important contributor to the construction of destination

descriptions (Section 5.1), allowing communicators to infer parts of their common

spatial knowledge.

The use of common knowledge as an important parameter of context im-

pacting on comprehension was explored in research on co-ordination games in

game theory. For example, in situations where two agents have limited means of

feedback, or lack means of direct communication, and concerted choice is sought,

people have shown high capacity of inference of the choice of the communication

partner. Typically, some entities at hand which have been known or assumed

to be known to the communication partner afforded themselves as natural co-

ordinating clues, also called focal points (Schelling, 1960). As Schelling (1960)

showed in his work on strategies in tacit coordination, objects or locations of high

uniqueness, that are part of common knowledge of the communicators and are

part of the communication domain, represented such focal points. These objects

and locations were then selected as references in situations of tacit inferential

communication. The example given by Schelling (1960) describes two soldiers

equipped with identical maps of the territory in which they had to meet, with no

means of communication. The most prominent or unique location shown on the

map presents a focal point, and is selected as the location where the likelihood

to meet is highest.

The research of Fussell and Krauss (1991), further confirmed by Lau and

Chiu (2001), shows that landmark knowledge estimates of other inhabitants of

a certain environment by their peers are accurate, although with a bias toward

one’s own knowledge. Such landmarks present focal points of the limited spec-

ified environment, e.g., a city. Furthermore, the differences between long-term

25


and short-term inhabitants are not major, which seems to be aligned with the

findings of Ishikawa and Montello (2006), about the relatively fast formation of

advanced forms of spatial knowledge. In this context, the five years long learning

period used to separate short-term from long-term inhabitants in the experi-

ments of Fussell and Krauss (1991) seems to be excessive. They draw, however,

a clear conclusion that the perceptions of the distribution of knowledge are so-

cially shared. Furthermore, one’s choice of characteristics and construction of

references in a referring expression is influenced by the perception of the hearer’s

familiarity with the described referent (Fussell and Krauss, 1992; Lau and Chiu,

2001).

In this thesis, the adjective familiar will be used for people with a-priori

spatial knowledge of the environment’s structure. The extent of this a-priori

knowledge is not known and has to be inferred by communication partners when

assessing the extent of common knowledge. People familiar with the environ-

ment can also be referred to as locals, a term which will be used in a restricted

sense, without its other social or regional connotations. The reasons why a rel-

atively limited amount of trips through a given environment suffices to judge an

individual a local will be presented later in the thesis.

2.2.5 Referential Communication

Referential communication is a type of communication where the speaker intends

to identify a referent to the hearer. In the most common case of referential

communication, the referent identified in the referring expression is unknown to

the hearer. A referring expression is defined by Dale (1992) as an expression

uniquely identifying a specific entity. Thus, a specific property of a referring

expression is that it has to be a distinguishing description of the entity referred

to. The speaker refers to a specific entity, member of a set of similar entities

(the context set), by singling out the distinguishing attributes (properties) of the

entity in mind. The combined role of the individual components of the referring

expression (the characteristic of the referent) is then to exclude the referent from

the context set, i.e. to single out the contrast set (Reiter and Dale, 1992). The

members of the contrast set are often called distractors.

Motivated by Grice’s conversational maxims, Dale provides a formalized

model to construct referring expressions, along with the Full Brevity Algorithm

(Dale, 1992), further improved in Reiter and Dale (1992). These algorithms con-

struct referring expressions that are the shortest description of a referent that is

still a distinguishing description of the referent.

26


Several properties common to human-generated referring expressions, namely

the adherence to the Gricean maxims, should be satisfied by automatically refer-

ring expressions (Dale and Reiter, 1995). Among the most important are:

• accuracy;

• brevity;

• incremental structure (attributes are ordered to rule-out as many distractors

as possible as early as possible);

• low redundancy, relevance

(Note that minimal possible redundancy may not be achievable, for com-

putational complexity reasons. Some redundancy occurs also in human

generated referring expressions, possibly as a function of the principle of

relevance.)

In their conclusions, Dale and Reiter (1995) note that an a-priori, explicit

implementation of the maxims is hardly possible, but very likely also unnecessary.

The outputs of their algorithms satisfied the maxims by being goal-driven. In

some sense, this result confirms the objections to the maxims as stated by the

relevance theory (Sperber and Wilson, 1986). The principle of relevance alone

is able to satisfy the characteristics required by the maxims and is also much

simpler to formalize. This approach is close to the approach presented in this

thesis.

Worth mentioning is previous research on referring expressions related to

place descriptions and route directions. The collaboration on references to ob-

jects, either mutually known and visually accessible by the recipient (Clark and

Wilkes-Gibbs, 1986; Heeman and Hirst, 1995), or unknown and inaccessible by

the recipient (Edmonds, 1994) was studied.

In this thesis, the case of descriptions of routes through coarsely familiar

environment is explored. These descriptions are a series of references to spatial

objects that are commonly known, but visually inaccessible at the time of the

communication of the directions. There is so far little research that specifically

looks into human route communication to wayfinders familiar with the environ-

ment.

2.2.6 Communication about Space

Place descriptions are expressions used to uniquely identify a place or a location

of an object in a restricted environment. In place descriptions, references to

27


elements of the city are introduced in a hierarchical order from general to highly

specific local references (Plumert et al., 1995b, 2001; Shanon, 1979). The answer

to the so-called where question varies with the context of the place description

(Tversky, 2003).

Route directions are verbal externalizations of mental representations of

routes, created during route planning and communicated by the speaker to the

hearer, i.e. the wayfinder. The goal of route directions is to unambiguously

describe the route to the hearer (Montello, 2005). As their goal is to uniquely

identify a place or a route, both place descriptions and route directions can be

perceived as spatial referring expressions.

The principles of route direction construction, their structure and the prin-

ciples of selection and inclusion of referents has been the object of research for

many years (e.g., Couclelis, 1996; Denis et al., 1999; Freundschuh et al., 1990;

Jarvella and Klein, 1982; Klein, 1979; Lovelace et al., 1999).

Route directions and place descriptions are constructed based on the spatial

knowledge of the speaker. The need to communicate a description of a specific

place or route to a wayfinder leads to a recall of this spatial knowledge, which

is selected and consequently externalized in form of a place description or route

directions. As noted by Couclelis (1996), the fine-level, detailed and embodied

route planning stage can be well formalized in computational models of navigation

and direction-giving. The result are so called turn-by-turn directions, typically

a series of references to streets using street names, connected by turn action

instructions. References to landmarks at decision points where turn actions occur

are more and more often included. The result are directions with a consistent

level of detail along the complete route. They are easily computed and provide

rich information to wayfinders with no previous environmental knowledge.

Turn-based route directions judged good by wayfinders in a series of exper-

iments are organized in an order which reflects wayfinder’s interaction with the

environment (Allen, 2000). The process of identification of locations along the

route that should be referred to in turn-based directions, and the identification of

appropriate environmental clues referred to was researched by Denis et al. (1999);

Lovelace et al. (1999); Michon and Denis (2001). References to salient features

along the route, mostly found at decision points (where turns occur in the route)

were found to be most useful in experiments (Lovelace et al., 1999; Michon and

Denis, 2001). An approach to automated identification of such salient features

has been presented by Raubal and Winter (2002), Winter et al. (2004), Notheg-

ger et al. (2004) and Elias (2003). They build on the research of Sorrows and

Hirtle (1999), studying the characteristics of landmarks, salient features of an en-

28


vironment. The salience of a feature in its environment is due to its uniqueness,

prominence or prototypicality. Thus, landmarks stand out due to their visual,

semantic or structural characteristics. These properties of landmarks impact on

their recall and establish a form of hierarchy in spatial knowledge (Taylor and

Tversky, 1992).

2.2.7 Directions and Cognitive Ergonomics

Turn-based directions of long or complex routes are difficult to remember. The

wayfinder needs to dedicate a part of the mental effort to remember the instruc-

tions communicated in the directions. This mental effort is combined with the

effort required for locomotion, for executing the primary task, e.g. driving, and

for preserving safety (including her or his own safety, that of possible passengers,

and that of other members of the traffic). Means for decreasing the complexity of

the communicated route directions have therefore been an important part of re-

search, combining aspects of human-computer interaction, communication theory

and spatial cognition, among others.

The communication of route directions may occur prior to the wayfinding

action, or in accompanying route directions during wayfinding. Furthermore,

route directions can be provided in a collaborative dialog, when the hearer may

provide feedback to the speaker, or in a one-directional communication. A model

of collaborative references in direction-giving dialogs was presented by Edmonds

(1994). For technical reasons, such as the reliability of natural language genera-

tion and speech recognition systems impacting on the performance of the feedback

recognition system, collaborative route directions systems are still rare beyond

research.

In unidirectional communication of route directions, other means of reducing

cognitive effort of the wayfinder have to be taken. One possible approach to

structured communication of route directions is that of providing references to

parts of the route incrementally. An instruction is communicated to the wayfinder

during locomotion, shortly prior to entering the place where a turn action has to

be taken (Maass, 1993). This is the solution of choice for current car navigation

systems.

Structuring the content is another simple means of reducing the cognitive

effort. Information grouped into chunks allows the information to be stored in

the short-term memory of people by adapting to its span (Cowan, 2001; Miller,

1956). Chunking is thus an effective method to adapt the information conveyed

to a recipient and to his or her short-term memory span (e.g., Klippel et al., 2003;

29


Richter, 2007a).

Dale et al. (2002, 2003) implemented a computational system providing hi-

erarchical route descriptions in an urban environment. Road status hierarchies,

road lengths and turn structures were used to construct a hierarchy of chunks

of instructions. The resulting directions were structured in a hierarchical tree-

like representation for use on mobile devices. While computationally simple, the

use of administrative street hierarchies may not necessarily be always cognitively

plausible. Furthermore, in contrast to human-generated route directions, Dale

et al. (2002, 2003) do not reduce the detail of the directions communicated. The

new structure is expressed through a graphical user interface, where the route is

still communicated in detail, turn-by-turn.

Changes in granularities of route descriptions due to the complexity of the

route have been also suggested by Agrawala and Stolte (2001). A graphical layout

system was designed to improve the communication of complex routes to users,

based on the structural properties of the route.

A large body of work on conceptualizations of the structural and functional

properties of routes, especially those of junctions, is presented by Klippel (2003a)

and Klippel et al. (2004). A set of primitive conceptual elements of route direc-

tions was identified and designed by wayfinding choremes. In consecutive works,

the analysis and use of the properties of routes to conceptualize route-specific

chunks of route information is demonstrated (Klippel et al., 2003; Richter, 2007a).

Klippel et al. (2003) use the term spatial chunking for the process of construction

of these route-specific route direction chunks. The following types of the spatial

chunking process were identified: landmark chunking, numerical chunking and

structure chunking.

Landmark chunking uses the presence of salient spatial features at the deci-

sion points where a turn occurs in the route to chunk the directions, e.g. “. . . turn

left after the church.”. This spatial chunking process proved to be the approach

of choice in the subject testing performed. The selection of the appropriate land-

mark for landmark chunking in route directions was further studied by Klippel

and Winter (2005). Numerical chunking consists of counting the number of de-

cision points where turns do not occur between two decision points where a turn

occurs along the route. This allows the creation of chunks of the following type:

“. . . turn left at the third intersection.”. Structure chunking relies on the unique-

ness of the spatial structure of the decision-points—intersections—where turns

occur, as illustrated by the example: “. . . turn left at the T junction.”. In a later

paper (Richter and Klippel, 2005), a fourth type of spatial chunking is identified,

based on the relationship of the route and line-landmarks. An example of this

30


type of spatial chunk is:“. . . follow the river.”. They suggest to extend the line-

landmark chunking approach with a stop criterion, e.g.:“. . . follow the river until

the bridge.”. The chunking processes may be applied to spatial chunks as such,

forming a hierarchy of chunks and higher-order chunks (e.g. “. . . turn left at the

next two T-junctions.”).

The experiment of Klippel et al. (2003) showed that spatial chunking is a pre-

ferred means to structure turn-based route directions by people for both a-priori

and accompanying route directions. Spatial chunking can be well operationalized

in computational algorithms and data structures (Hansen et al., 2006).

The spatial chunking approach relies completely on the structural and func-

tional properties of the route and its immediate environment for the construction

of the spatial chunk included in route directions. No information about the route

is avoided in the resulting route directions, making this approach usable also by

wayfinders with no a-priori knowledge of the environment. The resulting direc-

tions grow proportionally with the length and complexity of the route described.

While superior to traditional turn based directions, this approach may still lead

to the inclusion of excessive and thus irrelevant information to a wayfinder with

previous spatial knowledge of the city.

Finally, the reduction of the content communicated to only the pragmatically

necessary content offers itself as a means to reduce the cognitive workload of the

hearer. The pragmatic content of any message depends on the message itself (as

in classical communication theory), but also on the cognitive environment of the

hearer and of the speaker, and on the previous knowledge of the hearer (Grice,

1957; Sperber and Wilson, 1986). A schematic representation of the process of

communication and interpretation of route directions provided to a wayfinder

familiar with the environment is shown in Figure 2.3.

The knowledge of the hearer is inferred by the speaker. The existing spa-

tial knowledge of the hearer is used by the hearer to substitute the information

omitted in the place descriptions or route directions by the speaker for pragmatic

reasons. The speaker assumes that the hearer will be able to find her or his way

between the individual spatial features referenced in the route directions. Such

reduction of the content of route directions communicated is a method frequently

employed by people. It is, however, non-existent in current navigation systems.

Reduction is often in direct conflict with the requirement of providing certainty

that the information provided will be always usable.

The inspirational work of Frank (2003) provided the first insights into the

pragmatic information content in route directions, based on algebraic modeling.

An agent, i.e., a model of a human user, was represented as an algebra, and

31


(a) The speaker and the hearer have a-priori spatial knowledge of theenvironment, stored in their respective spatial mental representations(MR1 and MR2).

(b) The information communicated to the hearer is related to the spatialknowledge in the hearer’s spatial mental representation MR2.

Figure 2.3: Wayfinding in familiar environment. Information communicated inroute directions is related to the hearer’s existing spatial knowledge.

so was represented a decision context, too. As discussed by Frank (2003), the

interpretation of route directions by the hearer depends on the decision context

in which every direction is evaluated. His approach focused on the general prin-

ciple of comparing different direction algebras and comparing their pragmatic

information content in a given decision context.

One of the important aspects of context is the existing knowledge of the

hearer. Every reference adds to the operationalized knowledge of the hearer, and

thus the context changes with every reference included in the directions. Sim-

ilarly, the context changes with with the changing location of the hearer. This

allows tailoring of the selection of the information to be included in a message

to the hearer. In this thesis, a model of destination descriptions is presented,

aiming to adapt the pragmatic information content of a message communicated

to the wayfinder by reducing the quantity of references included in the destina-

tion descriptions. It takes into consideration the context in which references are

selected and relates it to the decision context in which the reference is used while

wayfinding.

Such destination descriptions do not reduce redundancy by avoiding the in-

clusion of references to common spatial knowledge that can be provided directly

by the hearer, but they do react better to changes of environmental conditions,

such as a changed traffic situation. As mentioned by Frank (2003), some route

32

M. TOMKO 2.3. MODELING AND FORMALIZATION

directions fail if a minimal error in the instructions is included, or if a deviation

from the route occurs. Destination descriptions are less prescriptive and allow

the wayfinder to change the route taken to a large extent. In contrast to previous

approaches to route directions, destination descriptions include references to mul-

tiple types of elements of the city, thus allowing variety in the resulting directions,

adapting the selection of the reference to the situation and thus maximizing the

relevance of the resulting destination descriptions.

2.3 Modeling and Formalization

2.3.1 Formalization and Functional Programming

Conceptual models present an abstract description of complex systems and phe-

nomena, with the focus on their intrinsic properties. To test the properties of

these models, their properties have to be formalized and implemented in an exe-

cutable or computational model.

When computationally implementing formal models of complex systems it is

necessary to ensure that the programmatic formulation of the formal system is

faithful, i.e. preserving the semantics of the formal definition. Ideally, an exe-

cutable mathematical notation should be sufficient for a computational validation

of the formal notation. Most current programming languages focus on efficiency

and performance, and their properties require the programmer to focus on lower-

level properties of the code, such as memory allocation. As a consequence they

do not enforce a programming style that would provide an unambiguous interpre-

tation of a formal model. In this regard, functional programming represented by

a group of programming languages such as Lisp, Miranda, Haskell and Scheme is

especially suited for scientific and research problems, with an increasing body of

industrial systems’ implementations. The high reliability of a code programmed

with a functional programming approach is, e.g., used by the telecommunication

company Ericsson. Their functional programming language Erlang is used to

manage mobile network communication. Traditionally, the focus of functional

programming languages is on formal strictness, resulting in a reliable code.

In functional programming, the specification of the model consists of func-

tions returning values based on the input arguments. Functions are defined over

a well specified range of types, and also return results of a well determined type.

This property can be checked before executing any function and is not dependent

on the values of the input or output. Complex functions can be composed from

elementary ones, where the output of one elementary function serves as input for

33


the next one. Vice versa, the result of complex functions can be computed by

atomic evaluation of its elementary functions. The behavior of the whole pro-

gram is transparent, as no external influences and special cases can be executed

without affecting all the inputs and outputs.

The functional programming language Haskell, and its precursor Gofer, were

previously successfully used for formalization of algebraic specifications of spatial

models, where semantic purity and strictness is necessary (Car, 1997; Frank, 2001,

2003; Frank et al., 2001; Kuhn and Raubal, 2003; Timpf and Kuhn, 2003). The

applications focused primarily on ontological and semantic modeling (Kuhn and

Raubal, 2003), granularity transformations in hierarchical data structures and

processes (Car, 1997; Timpf and Kuhn, 2003) and algebraic modelling of com-

munication of spatial information between cognitive agents (Frank, 2000, 2003).

The common requirement of these models is the explicit representation of the

semantics of the model in the algebraic representation, along with the possibility

to verify the validity of the resulting specification.

2.3.2 Modeling Complex Systems

Complex systems are defined by Simon (1962) as systems consisting of a large

number (i.e. individually unobservable) of parts that interact in a non-trivial

manner:

“In such systems, the whole is more than the sum of the parts

. . . given the properties of the parts, and the laws of their interactions,

it is not a trivial task to infer the properties of the whole.” (p. 468)

Systems, such as living organisms (systems of organs and tissues), the society

(the system of people and their social interactions) or urban systems (cities and

their constituents, such as the street network) are examples of complex systems.

Complex systems can be analyzed in a hierarchical manner, with the consid-

eration of the whole as a system of parts of finer granularity. The hierarchical

properties of these systems are frequently the result of an evolutionary process

(Bejan, 2000). Hierarchical properties grant complex systems a high level of ro-

bustness and efficiency. This can be related to the properties of the hierarchies

assisting in handling complex information in mental representations (cf. Sec-

tion 2.1). Decomposability, the fundamental property of complex systems, allows

the simplification of their descriptions across multiple granularities. At coarser

granularities of the description, only the agglomerative properties of the parts

have to be considered.

34


Two main types of descriptions are applied to the characterization of com-

plex systems: state descriptions and process descriptions (Simon, 1962). State

descriptions characterize the world as sensed, providing criteria for identifying or

modeling the elements of the system. Process descriptions characterize the sys-

tem as acted upon, by characterizing the actions necessary for the construction

of the system, or for the transition between individual granularities. Thus, both

models allow a more holistic picture of the system to be grasped.

This thesis presents a study of the state (what they consist of) and process

descriptions (how they are created) of destination descriptions. To provide the

process description of destination descriptions, the internal organization of the

representation of space they act upon must be described.

Granularity theory (Hobbs, 1985) provides an insight to the construction of

granular, hierarchical systems. Process descriptions of complex systems are built

upon relations of partial order, and the theory of granularity identifies processes

acting on the elements of the system which are at the cause of this partial order.

Network representations of complex systems allow the study of the struc-

tural properties resulting from the interactions between the parts of a system in

an abstract manner. Network analysis methods provide a plethora of measures

characterizing the role of individual elements of a system, their interactions and

the behaviour of a larger collection of network elements within the system (for

an overview, see Bocaletti et al., 2006; Hanneman and Riddle, 2005).

Network analysis and granularity theory provide means to describe the urban

systems in which the destination descriptions are generated. Network analysis is

a powerful tool to provide state descriptions of networked complex systems, while

granularity theory allows the processes contributing to the hierarchical properties

of those systems to be formalized.

2.3.3 Granularity

The processes in a complex, hierarchically organized system can be analysed by

studying the properties of its components and their interactions across multi-

ple granularities. The theory of granularity studies the principles of transition

between levels of complexity of a specific phenomena or model (Hobbs, 1985).

If, at a certain level of complexity of a system, two of its constituent phe-

nomena are indistinguishable from the perspective of the studied behaviour, the

system can be simplified by amalgamation or selection. It forms a system easier

to study, of a coarser granularity. Thus, granularity is the manifestation of the

relationship of indistinguishability.

35


The transition between individual granularities is based on rules of simplifi-

cation, consisting of operations executed upon the entities of the detailed set. In

the field of spatial information, these rules are amalgamation and selection (Stell

and Worboys, 1999). A hierarchical organization of a spatial dataset requires a

specification of a quantifiable property as a basis for the granular ordering. It is

important to emphasize that this research is concerned with semantic granularity,

depending on model simplification, as opposed to geometric granularity, so im-

portant for cartographic generalization. The semantics of an element considered

in the simplification relate to its functional importance in the structure.

The operations of simplification used determine whether the occurrence of

an element in a hierarchy is unique or finite (amalgamation) or repetitive (selec-

tion). During amalgamation, two or more elements of a set are merged into a

new element of coarser granularity. At coarser level, the constituent elements are

not present anymore. Selection, on the other hand, preserves the identity of the

selected element in coarser granularities, while other elements are left out. Fur-

thermore, the properties of the type of entities ordered determine the complexity

and the behaviour of the hierarchy. If the ordering is unique, the application

of amalgamation rules leads to hierarchical tree structures with fixed levels of

granularity. This is frequently the case of administrative hierarchies applied to

areal features, such as political land subdivision. It may not be the appropriate

ordering for other features, such as landmarks or streets (Alexander, 1988). For

instance, granular views of street networks will contain subsets of selected streets,

depending on a threshold value of a some property, but the identity of the streets

will not change.

Finally, the interdependence between the granularities of diverse types of

elements of the city is suggested by the characteristics of hierarchies in the spa-

tial mental representations. Thus, cognitively motivated hierarchical datasets

should allow for the transitions not only across granularities, but also between

the individual constituent types of element of the city form.

2.3.4 Spatial Modeling and Network Analysis

In complex systems, the structural characteristics of an element are largely deter-

mined by its relations to its peer elements. The distribution of the properties of

the individual elements are used to characterize the structural properties of the

whole system (Batagelj and Mrvar, 2006; Bocaletti et al., 2006).

In network analysis, networks are represented as graphs. Let G(V,E) denote

a graph G consisting of a set of vertices (vi, .., vn) from the set V (V 6= 0),

36


connected in pairs by edges e from the set of edges E. Let i, j and k be distinct

vertices of G. An edge e of the set E is then defined as an unordered pair of

vertices vi and vj: eij = (vi, vj). A subgraph G′ of the graph G is a subset of

vertices V ′ of the set V , and the subset of edges E ′ of the set of edges E connecting

the vertices in the set V ′. Any graph G can be represented in an adjacency matrix

A of dimensions V ×V with entries aij (i, j = 1, . . . V ) equal to 1 if an edge exists

between two vertices vi and vj, and 0 otherwise. This matrix is symmetrical.

The reachability of two vertices vi and vj is a central property of network

analysis. A path from vi to vj is a sequence of vertices and edges between vi

and vj, such that no vertex or edge is repeated more than once. A shortest

path between vi and vj is the path of least costs between the two vertices. The

cost function can be assigned to the vertices, edges, or both. The simplest cost

function counts the number of either vertices or edges visited along the path.

Algorithms such as Dijkstra (1959) and A∗ (Hart et al., 1968) can be used to

compute shortest paths between two vertices in a graph, based on an arbitrary

cost function. Note that the graph-theoretic use of the term path is different

from that introduced in Section 2.1.2. A route through an environment may be

represented in a graph by a path (in its graph-theoretical meaning).

Traditionally, the basic analytic structures in geographic network analysis

are graph representations of networks isomorphic with their geographic layout.

Linear geographic elements, such as streets elements, are represented in the graph

by edges e and intersections of these linear elements by vertices v. Such repre-

sentations are called primal graph representations. Common network analysis in

geographic information systems largely relies on the analysis of the metric prop-

erties of networks, where the cost function analysed over a network is related

to the length of an edge. Shortest path problems, travelling salesman problems

and other analytic tasks then require the assignment of costs to the edges of the

graph.

In the geographic realm, space syntax theory was one of the first to use a

dual graph representations of urban networks where graph vertices represent lin-

ear spatial features, and their intersections are represented by edges (Hillier and

Hanson, 1984). Such a representation emphasize the purely structural character-

istics of the urban network (Bera and Claramunt, 2003; Claramunt and Winter,

2007; Hillier and Hanson, 1984; Porta et al., 2006), and is therefore appropriate

for the analysis of the purely structural properties of the system.

Graphs analysis not only allows the analysis of network-like urban structures,

but also of spatial partitions of urban space. Partitions of space have been for

a long time studied with the help of specific types of planar graphs—Voronoi

37


diagrams and Delaunay triangulations. The dual structure of a planar Voronoi

diagram is the Delaunay triangulation (Okabe et al., 1999). Thus, Delaunay tri-

angulations are duals of Voronoi diagrams. Voronoi polygons are specially suited

to analyse neighbourhood problems in space, where the focal spatial feature is

represented as a vertex in the Delaunay triangulation and its immediate neigh-

bourhood is delimited by its dual Voronoi polygon. Voronoi diagrams have also

been studied across multiple granularities as a hierarchical data structure (Gold

and Angel, 2006). In a similar manner, Voronoi diagrams will be used in Chap-

ter 4.4 to study the partitioning of space by the reference regions of landmarks.

Note the difference in terminology and properties with the dual graph represen-

tations of networks, where dual-graph representations of linear graphs are not

homomorphic, i.e. by constructing the dual graph representation of a dual graph

representation, the primal graph representation is not obtained. The properties

are not symmetrical.

2.3.5 Basic Measures for Network Analysis

Among the most important structural properties of network elements are their

centrality characteristics. Three basic centrality measures allow the structural

significance of a network element to be characterized: degree centrality, closeness

centrality and betweenness centrality.

Degree centrality CDi of a vertex vi, in space syntax called connectivity, is a

measure specifying the number of direct neighbours (vertices connected by edges)

of a vertex:

CDi =

∑

j∈V

eij (2.1)

Degree centrality is a local measure, determining the characteristics of a vertex

only within the context of its direct neighbors.

Closeness centrality is a measure reflecting the average length of the shortest

paths from the vertex vi to all other nodes of the graph. Let pij denote the length

of the shortest path between the vertices vi and vj, and ni the number of such

shortest paths in the graph G:

CCi =

(ni − 1)∑

j=1..n pij

(2.2)

Nodes with high closeness centrality have low average length of the path to all

other nodes in the graph. When applied to a given urban network, this measure

reflects global properties of the structure of the city, revealing its core. In space

syntax this measure is known as global integration, or relative asymmetry, and

38


is applied on dual graph representations of the axial graph (Hillier and Hanson,

1984).

A localized measure of integration, considering only the network within a

radius of three steps is frequently used to reveal the variation of integration

across the network. This step-distance is based on empirical findings related to

the average length of pedestrian walks (Penn et al., 1998).

Betweenness centrality (also load, choice) quantifies the likelihood a graph

vertex lies on a shortest path between two other vertices of the graph. Between-

ness centrality CBi of the vertex vi was defined by Freeman (1977) as follows:

CBi =

∑

i6=j 6=k

njk(i)

njk

(2.3)

where njk(i) is the number of paths between vj and vk leading through vi. Be-

tweenness centrality provides a global value of a network element and thus allows

its structural characteristics to be compared with all other nodes in the graph.

Betweenness centrality is less influenced by the choice of the study area then local

characteristics of centrality.

Geographically embedded street networks or river systems display structural

patterns that can be studied by exploring the properties of their graph represen-

tations (Heinzle et al., 2006). The distribution of values of the different centrality

measures of graph elements reveals structural properties of the network and its

pattern. For example, regular grid-like street networks can be characterized in

terms of the distribution of the degree centrality and closeness centrality values

of the streets. The hierarchical model of space used in this thesis will be based

on the study of the distribution of betweenness centrality over several patterns

of urban networks to reveal their hierarchical structure.

2.3.6 Basic Elements of the Network

The selection of the elementary constituent of an urban network has an important

impact on the results of the analysis. Most geographers, urban analysts, and

therefore also traditional geographic information systems use street segments as

the building block of the street network.

Urban planners promoting the space syntax theory introduced the concept

of axial lines for urban space analysis (Hillier and Hanson, 1984). They define

axial lines as the longest line of sight in a convex space. The minimal graph of

axial lines, i.e. a graph with the smallest possible number of axial lines covering

all the convex spaces of the studied environment, is unique (Turner et al., 2005).

39


Axial lines have been studied as the basic building element of graph rep-

resentations for their relation to the conceptualization of convex spaces. It has

been, however, shown that further concatenations of street segments and/or axial

lines take place in human mental representations. Approaches based on strokes

or continuity lines have been proposed to account for such cognitive chunking of

linear elements of space (Figueiredo and Amorim, 2005; Thomson and Richard-

son, 1999). Concatenations of street segments sharing street names into named

streets was proposed by Jiang and Claramunt (2004), leveraging the semantic

properties of the analytical elements in combination with their structural proper-

ties. Such named streets then represent a functional modelling element of street

networks. Named streets are defined as the set street segments sharing a label

(street name), and are represented as a single modelling element of the network

analysed. Named streets are further used as the basic modelling element of net-

works in this thesis.

40

Chapter 3

Destination Descriptions

In this chapter, the concept of destination descriptions is introduced, as well as

their basic characteristics. These properties are derived from empirical experience

and illustrated with example scenarios.

3.1 Scenario

Let us return to the example of Stephanie, an inhabitant of Hannover return-

ing home from a business trip, as presented in the introduction of this thesis

(Section 1.1). In the given situation the two communication partners can make

several assumptions. The taxi driver can make at least the following inferences

about Stephanie:

• Stephanie wants to get to the centre of Hannover, not another city, nor

the centre of a local village. If she wanted to get to these places—unusual,

although possible destinations for a taxi driver at the Hannover airport, she

would specify them in more detail. Furthermore, the whole set of directions

enforces the assumption (the centre combined with Staatstheater).

• Stephanie is trying to be collaborative, and provides optimal directions in

the given situation. This inference is simple to make, as it is Stephanie’s

interest to get home as soon as possible (at least due to taxi costs). Longer

directions may not only not be necessary (the driver will find the way), but

restricting the driver to a specific route removes the driver’s options to op-

timize the route with his expert knowledge (e.g. in case of changes in traffic

conditions). Finally, this could be interpreted as patronizing behaviour.

• Stephanie wants to be driven to her destination in the cab she is sitting in,

i.e. she does not want to change modes of transport midway. Thus, the

road network available to cars (taxis) will be used.

41

CHAPTER 3. DESTINATION DESCRIPTIONS

• Stephanie has some knowledge of Hannover (i.e., either from a map, or she

may even be an inhabitant of Hannover). This inference may be based on

the fact that she knows the centre (a vague concept, someone with no prior

knowledge of a city tends to use official references), knows that Luisenstrasse

is a laneway, and knows a landmark in its vicinity. Finally, after the first

unsuccessful utterance an unfamiliar visitor of Hannover would ask the taxi

driver to find his own way with a map.

On the other hand, Stephanie can assume at least the following in the given

situation:

• The taxi driver knows Hannover (i.e., has previous knowledge acquired by

driving around the city. This may include, e.g. the capability to drive out

of the airport, and reach the city.).

• As they have never met, the taxi driver does not know her house (thus, she

has to provide some route directions).

• The taxi driver knows the most prominent places, landmarks and streets of

Hannover. She also assumes that the taxi driver knows what she means by

centre and Staatstheater.

• The taxi driver may not know all the streets in Hannover (the street name

and number would be otherwise sufficient). This is confirmed by the un-

successful utterance of Luisenstrasse.

• The taxi driver will make an effort to understand and interpret her direc-

tions correctly (i.e. will behave collaboratively, as he wants to earn his

money, which he only gets if he gets to the correct destination).

• The taxi driver will use the cab to drive her home. Thus, route directions

provided should be optimized for a car driver (and not e.g. a helicopter

pilot, which would need only a set of coordinates).

The assumptions inferred from the situation at hand may be coarser or more

detailed. Some may even seem far-fetched and ridiculous. They demonstrate,

however, the richness of background information that may be used to interpret

the directions given and that is not uttered explicitly. Difficulties in interpretation

may arise, but the context brought by the remaining references clarifies them.

Should the taxi driver know a prominent street Luisenstrasse elsewhere in the

city, the reference to the centre and the Staatstheater make it obvious that it

is not the correct interpretation of the reference. Should he not know such a

42

M. TOMKO 3.2. DEFINITION OF DESTINATION DESCRIPTIONS

street at all, he may infer from the directions that the street is right next to

the Staatstheater, and will wait for environmental clues in its vicinity to see

whether he needs more information. Of course, in case that the taxi driver knows

the Luisenstrasse in the centre of the city, the additional references may seem

redundant. The role of the additional references is to construct a full referring

expression, uniquely identifying the destination, and thus to reduce ambiguity.

The type of route directions provided by Stephanie to the taxi driver will

be from now on called destination descriptions. A more formal definition of the

destination descriptions, along with their main properties, follows.

3.2 Definition of Destination Descriptions

In this thesis, destination descriptions in urban environment are defined as fol-

lows:

Definition 6 Destination Descriptions (in urban environment): Destination de-

scriptions are a referring expression uniquely describing a destination of a route

in a given urban environment, consisting of a hierarchically ordered set of ref-

erences to prominent spatial features of various types, provided in the context of

inferential communication to a hearer with assumed a-priori spatial knowledge of

the environment.

Destination descriptions are closely related to place descriptions and route

directions. They combine the characteristics of place descriptions, with the added

constraint to adapt the description to the context of the route. As such, desti-

nation descriptions focus on the specification of the destination, thus giving the

answer to the Where? question, as opposed to turn-based route directions, fo-

cusing on providing information about How? to get to the destination. In the

example above, Stephanie provides the taxi-driver with a description of the vicin-

ity of the destination, as opposed to the turn-based directions as provided, e.g.,

by Google Maps (Figure 1.1) providing full information on the actions to take to

reach the destination. In destination descriptions, the speaker assumes that the

hearer is able to fill this information from a-priori spatial knowledge.

43


3.3 Structure of Communication of Route Di-

rections

Consider Figure 3.1, which presents three templates of the structure of the route

directions communication process. Figure 3.1a schematically depicts turn-by-turn

directions between the start and the destination (TBT ). Such route directions

contain references to to streets, connected with action directions at decision points

along the route. This is in contrast to destination descriptions, as depicted on

Figure 3.1b. A sequence of references identifies the destination of the route.

Destination descriptions (DD) are communicated with the context of the start in

mind (either explicitly requested by the hearer, or inferred in case of co-location of

the speaker and the hearer). The references included in destination descriptions

do not cover the whole route. In fact, the route is never communicated to the

hearer. Thus, the first reference of destination descriptions relates to a spatial

feature in the proximity of the destination, and not in the proximity of the start of

the route. The speaker assumes that the references made are part of the common

knowledge, shared with the hearer.

(a) Turn-based directions (TBT)

(b) Destination description (DD)

(c) Destination description followed by a seriesof turn-based directions

Figure 3.1: Structure of navigation instructions from start to destination of aroute (see text for more details).

The structures of communication of route directions can be merged to provide

a generic structure of route directions for a hearer with a-priori spatial knowledge

(Figure 3.1c). Destination descriptions continue until they reach a level of detail

where either the speaker assumes no further reference is part of the a-priori spatial

knowledge of the hearer, or the hearer provides feedback when unable to identify

the reference in her or his a-priori spatial knowledge. The speaker then changes

the form of communication of route directions from destination descriptions to

44

M. TOMKO 3.4. SELECTION OF REFERENTS

turn-by-turn directions, providing detailed instructions about how to reach the

destination. Until then, the references include in destination descriptions are

non-prescriptive, i.e. do not require the hearer to follow an specific route to

the destination. Thus, although references to streets are made in the example

of Stephanie and the taxi-driver (“(Luisenstrasse) . . . is off Rathenausstrasse.”),

the taxi-driver is free to come from a different direction and uses the references

only as a description of the destination’s vicinity.

The transition to turn-based directions may happen earlier or later in the de-

scription, depending on the extent of the inferred hearer’s a-priori spatial knowl-

edge. If the wayfinder is highly familiar with the environment and thus her or his

a-priori spatial knowledge is rich, the sequence of turn-based route directions pro-

viding detailed information about the route to take is not present in the resulting

route directions.

The generic structure of communication of route directions can be extended

for routes with several destinations, such as in the travelling salesman problem.

As shown in Figure 3.2, the structure in Figure 3.1c can be used repeatedly to

produce route directions for multiple destinations (i.e., Dest.’,Dest.).

Figure 3.2: Structure of spatial communication for a route with a transition point(Dest.’ )(see text for details).

3.4 Selection of Referents in Destination De-

scriptions

In an ideal case where the destination of the route is known to the speaker and the

hearer, and this knowledge is also mutually known, the speaker may use a direct

reference. It is, however, almost impossible to establish such mutual knowledge

(Section 2.2.4). Furthermore, for most locations of a particular city, such mutual

knowledge is also highly unlikely. Only members of a narrow group of people with

large overlaps in common knowledge (e.g. family members, colleagues, fellow

students or members of a social club), have means to infer the mutuality of their

spatial knowledge.

Thus, the speaker includes also references to entities of coarser granularities

in the destination descriptions, to construct a spatial referring expression and

thus uniquely identify the referent to the hearer. Among multiple alternatives,

45


the reference to the most prominent feature provides highest likelihood of correct

interpretation (Pattabhiraman and Cercone, 1990). It requires the least cognitive

effort from the speaker to be retrieved from the set of alternatives. Furthermore,

the speaker assumes that a prominent referent will require a low cognitive effort

of the hearer when linking the references made by the speaker to his or her spatial

mental representations of the same referents.

So, the speaker refers to well-known elements of the city in the proximity

or containing the destination. With every additional reference made, the spa-

tial extent in consideration is more restricted. Consecutive references point to

entities within the extent of the previous referent. Every reference specifies the

destination in more detail, as well as mutually enforcing the interpretation of

the other references. The selection of consecutive references by the speaker is

made in a manner maximally exploiting the meaning of the previous reference,

thus maximizing its relevance. Furthermore, the understanding of every consecu-

tive reference is achieved in the enriched context of which the previous reference

is part. The combination of the references uniquely describes the destination,

and any omission of a reference would increase the ambiguity of the resulting

directions.

The hearer expects a reference to a referent of maximal possible relevance in

a given context. If multiple interpretations of a reference are possible, the optimal

one is the one maximizing the relevance of the utterance (Hasida et al., 1995).

The cognitive environment (Sperber and Wilson, 1986) of the hearer changes

with every reference made. Every reference enriches this environment and makes

it less ambiguous.

The destination description provided by Stephanie in the example above

can be represented by a schema of the referents included (Figure 3.3). The

hierarchy of granular references reflects their prominence within Hannover. The

prominence of a spatial feature drives the selection of referents for destination

descriptions. The cognitive effort to link the reference to its mental representation

is lower among prominent entities. Thus, by applying the principle of relevance,

prominent spatial features are more likely to be included in route directions.

Let us explore the selection of the referents more formally. Let us call the spa-

tial mental representation of the speaker S and that of the hearer H (Figure 3.4).

The content of S and H are the mental representations of the experienced ele-

ments of the city, i.e. the reality R. Thus, S and H are sets of these mental

representations, and they are incomplete representations of R. The perception of

the environment by an agent is subjective, incomplete, and determined by per-

sonal characteristics of the agent, and therefore the content of the sets S and H

46


Figure 3.3: Schema of the hierarchy of references in the destination descriptionfor the route from Hannover Airport to Luisenstrasse.

are not equal (S 6= H).

Figure 3.4: Mental representations of the speaker (S) and the hearer (H), formedby experiencing the reality. Common knowledge of the speaker and the hearer(C) represents the subset of mental representations of elements of the city sharedin S and H. Note that the hierarchical structures of S, H and C are representedas tree-like structures for simplicity only (Arc denote the partial order in thehierarchy).

S and H, consist of mental representations of elements e1..n of any of Lynch’s

five types of elements of the city. The elements in S and H are organized in hierar-

chical structures defined by the respective partial orders hS(S,≤) and hH(H,≤).

The structures hS and hH are different, as they are results of individual spatial

experience of the reality R (hS 6= hH).

To be able to infer the meaning of the route directions correctly, the structure

of the hearer’s spatial knowledge must be compatible with that of the speaker. As

the familiarity of a person with a certain environment increases, the person’s spa-

tial mental representation increases in size and the hierarchical structure emerges.

47


If the spatial behaviour of the speaker and the hearer has common characteristics,

as is the case with most people inhabiting a certain area, the structures hS and hH

share some common characteristics. The intersection of the sets S and H is the

set of elements forming the common knowledge C of the speaker and the hearer

(S ∩ H = C). The common knowledge C has a hierarchical structure hC(C,≤).

For any two elements ea and eb, parts of common knowledge C (ea, eb ∈ S,H,C),

the following holds true {ea, eb|if ea ≤S eb and ea ≤H eb then ea ≤C eb} . In case

that the set of elements in C is large, i.e. the size of C approaches that of S, the

speaker can provide destination descriptions to the hearer in an inferential spatial

communication, and the likelihood of the hearer inferring the meaning intended

by the speaker is high.

The following cases may then happen in the communication:

1. The reference provided by the speaker is to an element ex that is part of the

common knowledge (ex ∈ C). Assuming that there are no linguistic barriers

(the reference to the referent is made so that the meaning interpreted by

the hearer is to the same referent), understanding is achieved.

Figure 3.5: Detail of the hierarchical structures S, H and C (see text for details).

Imagine that the referent is to the element e5 in Figure 3.5 (detail of Fig-

ure 3.4), and that the reference to this referent is unambiguous (e.g., it is a

castle, and there is only one castle in the set C). If the castle was also the

destination of the route described in the destination description, the direc-

tions: “Go to the castle.” will be correctly understood by the hearer. This

is a frequent case in the communication with taxi drivers where the intended

destination is a well known, prominent location with an unambiguous and

familiar name.

2. The reference provided by the speaker is to an element ex that is not com-

monly known ((ex ∈ S) ∧ (ex /∈ H)). In such a case, a coarser referent in

the spatial vicinity of the destination should be selected by the speaker.

48


This can be demonstrated by a small alteration of the above scenario. As-

sume that the area of the castle consists of the Old Palace (element e10)

and the New Palace (element e11) (Figure 3.5). If the speaker refers to e11,

which is not known to the hearer, understanding is not be reached. Then,

the speaker should provide the reference of coarser granularity (e.g. e5, the

Castle).

Note that the semantics of the reference may allow the hearer to infer the

meaning of the reference: a palace is a part of the Castle; once there, the

driver may ask for detailed turn-based directions, or find the palace based

on environmental clues. If it is not possible to provide references of coarser

granularity, turn-based directions of full detail must be provided.

An extension of this scenario can be imagined. Imagine that the set C con-

tains elements that can be referred to with the same reference (e.g., a name). A

simple, direct reference to one of these entities is then ambiguous in the spatial

context in which the destination description is provided. Destination descrip-

tions are referring expressions, as they must identify the destination of a route

uniquely. Incomplete referring expressions lead to a lack of orientation (Paraboni

and van Deemeter, 2002). The principle of relevance drives the interpretation of

the references—the cognitive environment in which the references are provided

contains also parameters of spatial context. The current location assists with the

hearer’s interpretation of the reference.

This reasoning is closely related to the interpretation of meanings in collab-

orative games, as introduced by Hasida et al. (1995) and Hasida (1996). If a

communicator sends a message m1, and this message may encode the contents

c1 and c2, and there is a possibility of the message m2 to convey the content c2

uniquely, then the most relevant interpretation of a message m1 corresponds to

the content c1 (Figure 3.6). Note that Hasida et al. (1995) provide the reasoning in

terms of Grice’s cooperative principle, as opposed to the interpretation provided

here, which follows the more cognitively plausible relevance theory of communi-

cation. The implications for the selection of references and the interpretation of

their meaning are illustrated in the following examples.

Let us assume that there are numerous castles in the proximity of the location

in which the communication of the route directions occurs, all present in the

common spatial knowledge C (Figure 3.7). Furthermore, let us assume that the

directions are provided by the speaker to the hearer in the spatial context of

the start location e13. The destination castle where the speaker wishes to get is

represented by e5. Several scenarios are then possible:

49


(a) Possible encoding of con-tent by messages

(b) Relevance-based inter-pretation of the encoding

Figure 3.6: Possible encodings of two contents in messages, and their interpreta-tion based on relevance theory.

Figure 3.7: Detail of the hierarchical structure of the set C (see text for details).

1. A second castle is found in the proximity, represented by the element e9.

The speaker and the hearer are in the spatial context of e13. The directions:

“To the Castle!” will be interpreted by the hearer to mean: “to e5”, as pre-

dicted by the principle or relevance. The element e9 is of lower prominence

then e5, requiring more cognitive effort to relate to the reference made. The

meaning conveyed by the stimulus is interpreted in a manner to maximize

the relevance of the stimulus.

2. The current location of the speaker and the hearer (e13) is within a second

castle (element e6). Note that both elements e5 and e6 have equivalent

granularity in the structure of the space concerned, i.e. they have similar

prominence. Physical co-presence is a strong determinant for the interpreta-

tion of references. The principle of relevance will lead the hearer to interpret

the reference to the castle as meaning the element e6. The relevance of the

directions: “To the Castle!” will be judged low by the hearer, as a location

matching this reference has already been reached. The meaning intended

by the speaker will not be the meaning understood by the hearer. A refer-

ence to a coarser element of C therefore needs to be added, such as e2. The

resulting directions will then consists of two references (to e2 (Prague) and

50

M. TOMKO 3.5. COMMON KNOWLEDGE

e5 (castle)), such as: “the Prague Castle”.

3. The destination is located close to multiple elements of similar names and

prominence. Imagine three neighboring, equidistant castles of similar promi-

nence e4, e5 and e6. The castles e4 and e5 have the same distance from the

current location e13. The interpretation of any direct reference to e4 or e5

would be ambiguous, and at the granularity of e2 both castles are indistin-

guishable. Destination descriptions to the element e2 should be provided,

and from there, turn-based directions are necessary. Note that we assume

that the reference to both elements is the Castle. Should it not be so, the

proper noun, or a referring expression based on characteristic of the castle

can be used: the Old Castle, the Prague Castle.

3.5 Common Spatial Knowledge of the Environ-

ment

Destination descriptions require the hearer to use a-priori spatial knowledge in

combination with the content explicitly communicated in destination descriptions

through references. The meaning of the references has to be interpreted, and this

interpretation requires the consideration of the context in which the reference is

made. These characteristics fundamentally differentiate destination descriptions

from turn-based directions, where neither a-priori spatial knowledge, nor the

contexts in which the directions are generated are required to properly interpret

the meaning of the directions and reach the destination.

A-priori spatial knowledge required for successful communication or inter-

pretation of destination descriptions has all the properties discussed earlier: an

experiential and integrated hierarchical structure. As the speaker provides infor-

mation about the destination of the route, he or she must always have a-priori

spatial knowledge of the environment.

On the other hand, the existence of a-priori spatial knowledge of the hearer

must be inferred by the speaker from the perceivable characteristics of the hearer

and other stimuli in the cognitive environment, defining the communication con-

text. The hearer relates the directions received to her or his own spatial knowl-

edge, as shown in Figure 2.3.

It is common to qualify people living in and knowledgeable of certain lo-

cations as locals. The term local will be used in this thesis as alternative for

agent with a-priori knowledge of the environment. An overlap in the mental

representations of the speaker and the hearer (the domains), as well as in their

51


hierarchical structures is assumed in this thesis. The coverage of their common

spatial knowledge enables inferential communication of route directions in a given

urban environment. In the communication situation of a navigation system and

a human user, the system has to infer that the wayfinder has some a-priori spa-

tial knowledge of the environment. The references included in the destination

descriptions have to be selected in a manner maximizing the reliability of the

resulting destination descriptions.

The hearer makes an effort to interpret the destination descriptions in a man-

ner that will maximize their relevance in conjunction with the current context and

her or his spatial knowledge. The principle of relevance is the driving force guid-

ing the interpretation of the destination descriptions. To reach consensus between

the speaker and the hearer, the reference selected by the speaker is communicated

in a destination description and related to the most relevant entity in the spatial

mental representation of the hearer. In case of unambiguous interpretation of the

reference by the hearer, the related spatial mental representation refers to the

same entity in the real world. This is only possible if the speaker referred to an

entity that is part of the common knowledge of the two communicators.

In the example above, Stephanie communicates a destination description to

the taxi driver in an inferential communication process. Stephanie and the taxi

driver are mutual strangers. They mutually assume at least coarse familiarity of

the communication partner with Hannover, and thus the existence of common

knowledge. This assumption is based on inference from the respective cognitive

environments of the two communicators.

Supportive evidence (e.g., sitting in a cab, current location) allow them to

tacitly assume a shared functional perspective on the urban structure, determined

by the ostensive choice of the mode of transport and the starting point to which

the directions are related. With this tacit evidence, the hearer is able to reach

understanding about the meaning of the references selected by Stephanie for the

destination description.

In these references, Stephanie’s spatial knowledge is disclosed ostentatively

to the taxi driver. The hearer discloses the shared possession of this knowledge by

the act of driving, without questioning the content of the destination description.

Similarly, in the first communication attempt, the taxi-driver disclosed the lack of

spatial knowledge by not driving to Luisenstrasse directly after the reference to

this street. It is thus possible to extend the generic model of route directions with

the case of communication of route directions in a dialog, where a negotiation

about references may be included at any stage (Figure 3.8). Ambiguity resolution

may also be implemented in a computational system as a detailed, hierarchically

52

M. TOMKO 3.6. CHARACTERISTICS OF DESTINATION DESCRIPTIONS

finer, set of turn-based route directions included upon request by the user.

Figure 3.8: Structure of spatial communication with a negotiation dialog (Neg.).

Note that incorrect interpretation of the references is possible, and osten-

tatively manifested by the driver driving to a different destination than the one

meant by the speaker. Such situations can be disambiguated by the sequence

of multiple references, where each consecutive reference specifies the destination,

as well as the previous references, in more detail. The combination of the refer-

ences is highly unique in a given communication context. As such, a full referring

expression is composed.

3.6 Characteristics of Destination Descriptions

Destination descriptions represent a specific case of referring expressions as de-

fined by Dale (1992). They serve the goal of uniquely identifying the destination

of a route to a wayfinder with at least coarse previous spatial knowledge of the

environment in question. They do so without prescribing the detailed route to

be taken, leaving the freedom of choice and the possibility of alteration to the

wayfinder during locomotion itself. Furthermore, the brevity of destination de-

scriptions is a significant property lowering the effort necessary to remember them

and so allows the wayfinder to concentrate the cognitive effort on other tasks, such

as wayfinding or driving.

In destination descriptions, references are serialized hierarchically, usually

in order from references to best known, most prominent referents in the wider

vicinity of the destination or in its coarse direction, to more detailed references

to spatial features in the close vicinity of the destination. The ordering of the ref-

erences in the actual verbalization may, however, differ due to linguistic reasons.

Note that the change in granularity of references selected is due to the narrowing

of the space within which the destination has to be singled out and not to the

structure of the route as such. Changes in granularities of route descriptions due

to the complexity of the route have been previously suggested by Agrawala and

Stolte (2001), Klippel et al. (2003) and Richter (2007a), among others.

As each reference provides information about a narrower section of the space

described, even descriptions of long and complex routes are relatively brief. The

presence of references to a variety of types of spatial features allows an efficient

53


and brief description of the route. As the directions proceed from general to more

detailed references, the certainty of the speaker that the hearer has sufficient

spatial knowledge decreases. In human communication, speakers then change

from destination descriptions to turn-based directions.

The examples presented show how the current location and the granular

structure of the environment impact on the selection of references by the speaker,

and consecutively their interpretation by the hearer. The minimum spatial detail

communicated in a reference occurring in destination descriptions is that of the

first reference. The selection of consecutive references then consists of the task of

retrieving a relevant reference within the context area specified by the previous

reference.

The examples presented point to the importance of a cognitively motivated

model of the space, in order to be able to execute a formal model of destina-

tion descriptions computationally. Therefore, before introducing the model of

destination descriptions, the cognitively motivated, experiential structure of the

environment will be discussed.

54

Chapter 4

Hierarchical Data Structures for

Destination Descriptions

In this chapter, the hierarchical organization of the urban environment is ex-

plored, in order to derive means of structuring spatial datasets hierarchically for

use with the model of destination descriptions.

4.1 Hierarchical Structure of the Urban Envi-

ronment

The urban environment consists of various spatial features, such as suburbs,

prominent landmarks, streets and their junctions in the street network, water

canals and city walls, to name a few. All and any of those may and do appear in

human generated route directions.

These spatial features may be hierarchically organized in individual hierar-

chies, a fact often used in formalized classification systems, such as administrative

hierarchies of land subdivisions or the classification of streets in urban networks

by number of lanes and other designer criteria. Note that this organization is

only conceptual and not perceivable in reality.

As mentioned, administrative classifications are not necessarily suitable for

cognitively ergonomic provision of route directions. This is due to several factors:

Classification criteria Administrative partition of space is motivated by other

then experiential criteria, and is thus not necessarily compatible with the

experiential classification of urban space in spatial mental hierarchical repre-

sentations of locals. Similarly, the classification of streets in street networks

may not correlate well with the experience of prominence of streets in the

city.

55

CHAPTER 4. HIERARCHICAL DATA STRUCTURES

Structure Administrative hierarchical classifications are designed to provide a

finite number of hierarchically ordered classes of the features classified, de-

termined in advance. They form a levelled hierarchy that has a partial

order between its classes. Such discrete levels are not necessarily reflected

in our mental representations. An element may appear through multiple

levels of granularity. Imagine a landmark, such as the Eiffel Tower—it can

well serve as a global landmark embodying the whole of Paris, as well as a

local navigation aid for a wayfinder in its vicinity.

Homogeneity Administrative classifications of spatial features are usually ho-

mogeneous. Only one type of Lynch’s elements of the city is classified in

the hierarchy. The interdependent inclusion of spatial features from sev-

eral of such experiential hierarchies in destination descriptions suggests a

single, integrated hierarchical structure. Thus, the relationships between

the individual hierarchies of different types need to be established for inter-

changeable provision of destination descriptions.

As discussed later in this section, prominence is the principal factor influenc-

ing the structure of experiential hierarchies. Thus, in order to computationally

implement the model of destination descriptions presented, spatial data struc-

tured along cognitively motivated principles are required. The consideration of

the prominence of the spatial features that can be selected as references is an

essential factor influencing the composition of the destination descriptions.

The transition between the level of detail of references in destination descrip-

tions is often accompanied with a change of the type of the referent. At a certain

granularity, people make the choice of the referent most appropriate to the spatial

structure of the route described. The concept of integrated spatial hierarchies and

their construction is introduced in more detail in Section 4.5.

4.2 Experiential Hierarchies

Experiential hierarchies form in mental representations as a product of the in-

teraction of wayfinders with the environment. The intensity of experience of

a spatial feature is related to its functional, structural or semantic (individual)

prominence in a specific environment. This experience of prominence establishes

a partial order between the mental representations of the individual spatial fea-

tures, and an experiential hierarchy emerges. People’s individual experiential

hierarchies represent one of the fundamental structures on which they base their

assumptions about the spatial knowledge of others (Fussell and Krauss, 1992).

56

M. TOMKO 4.2. EXPERIENTIAL HIERARCHIES

Individual experiential hierarchies cannot be externalized. Common properties

of such structures among locals—overlaps in the content and hierarchical orga-

nization of common spatial knowledge—warrant the success of inferential com-

munication. Common knowledge of a spatial feature may be inferred, and the

selection of references to elements of a city in communication such as destination

descriptions is possible. Thus, the construction of hierarchical datasets following

cognitively motivated principles will provide means to draw qualified estimates of

the relative prominence of spatial features in a given environment. For simplicity,

such hierarchical orderings of spatial datasets are called experiential hierarchies,

too.

The prominence of a spatial feature influences its rank in hierarchical mental

representations. Two properties must be met by a spatial feature to be included

in destination descriptions:

Prominence The feature must stand out from the background, be salient, dis-

tinct, or unique (Sorrows and Hirtle, 1999).

Identity To be included in destination descriptions, the feature must also be

uniquely identifiable. If this condition is not met, it is difficult for the

speaker to select it as an unambiguous reference.

The prominence of each of the types of elements of the city is the result of

its visual, semantic and structural characteristics. For each of these types, how-

ever, the relevance of these characteristics is different. Districts are difficult to

be perceived from a single point. They are experienced as an cohesive, homo-

geneous environment, sharing characteristics and distinct from its surrounding.

The semantic and structural characteristics are therefore comparatively stronger

then visual ones. Paths may be experienced due to their structural properties,

facilitating frequent trips through the city due to their structural embedding.

Landmarks are remembered due to their unique visual or semantic properties,

such as the distinct characteristics of their facades or the type of business resid-

ing in a building.

The identifiability of the feature is an important parameter in communica-

tion among mutual strangers. If two locals are not mutual strangers (i.e., are

member of a community, such as a family), their common spatial knowledge has

the properties of mutual knowledge. This extended, shared context allows for

descriptive reference in destination descriptions, such as: “Go to the cafe where

we celebrated your birthday last year”. Among strangers, such references are

not possible. Thus, the speaker has to rely on a common system of labelling of

spatial features, such as street names, suburb names, and well-known vernacular

57


toponyms. In this model, we will rely on administrative toponyms, such as suburb

names and street names. Note that the entities labeled by those toponyms may

not necessarily have the exact extent as specified in the administrative specifica-

tions. The hearer does not need to know the exact specification of the boundaries

of Hannover to understand the reference to it.

4.3 Composition of the Urban Environment

As introduced in Section 2.1.2, five elements of the city form (concepts) are

recognized as the basic building blocks of mental representations of urban space.

In the model of destination descriptions presented later, three of those elements

are explored, namely districts, landmarks and paths. Due to their vague definition

and operationalization in the work of Lynch (1960) (for a different approach to

re-definition of the types of elements of the city, see Conroy Dalton and Bafna,

2003), their refined definitions as used in this thesis are introduced:

Path: a path is defined as a one-dimensional conceptual entity of the environ-

ment along which observers move. In this thesis, the city is experienced

through movement along streets, forming the street network. The streets

are then the physical entity along which the wayfinders move. Note the

distinction with the use of path as a graph-theoretical term.

The basic element of this network is the named street (further used inter-

changeably with street), consisting of all the street segments sharing the

same name (label). Reasons for this choice are explained later in this chap-

ter.

District: a section of the city with a two-dimensional extent, sharing some com-

mon characteristics and have a distinct inside and outside.

Landmark: Landmarks are point-like spatial features, serving as spatial refer-

ences standing out from their environment.

Two types of districts are explored in our model: the hierarchical partition of

space into suburbs (administrative partition), and a more cognitively motivated

hierarchy of spatial partitions, created by reconstructing the reference regions of

landmarks.

The two remaining types of elements of the city, nodes and edges, are not

formally explored in the remainder of this thesis. The extension of the model to

include these elements is a subject for future work.

58

M. TOMKO 4.4. HIERARCHIZATION OF ELEMENTS OF THE CITY

4.4 Hierarchization of Elements of the City

Complex systems, such as the urban structure, may be analysed at multiple

granularities. The definition of the elementary block, the finest granularity of

the system, is an arbitrary choice of the analyst, motivated by the characteristics

of the phenomena studied. In this Section, the principles of hierarchisation of

the three types of elements of the city are introduced, as used in the model of

destination descriptions. In order to rank elements in a hierarchy, a parameter

determining the partial order of the elements is determined. Furthermore, the

integration of these diverse hierarchies is explored.

4.4.1 Hierarchies of Landmarks1

A landmark is prominent due to its visual, structural or semantic properties. It is

prominent only as far as it is either unique, or highly salient in the specific spatial

context—the reference region of the landmark. The landmark is the anchor of

its reference regions (Kettani and Moulin, 1999). In its reference region, the

landmark is dominant, i.e. it is the most prominent element of the region. This

determines the reference region as the region of prominence of a landmark, and is

a parameter for establishing a partial order—hierarchical ranking—of landmarks.

The example given by Montello et al. (2003): “The area around Eiffel Tower.”

suggests that a landmark (the Eiffel Tower) is a generator of a cognitive represen-

tation of a region in its vicinity. The region may be of the size of its vista space,

as it is the case of the local 7-Eleven, but may be even larger. The uniqueness of

the Eiffel Tower in its visual, semantic and structural parameters grants it a ref-

erence region of minimally the size of Paris. Furthermore, in its reference region,

a landmark has the properties of a focal point, a natural focus of the given region

toward which attention in the given spatial context will gravitate, for instance in

tacit communication (Schelling, 1960).

Lynch distinguishes between distant landmarks, physically inaccessible, and

visible from a large region, and local landmarks, physically and visually accessible

in a restricted space. Therefore, there may be several levels of reference regions

of the landmarks. The properties of the reference regions will be further explored

as an input for integration of district and landmark hierarchies.

1Parts of this section were previously published in (Winter et al., to appear 2007). I acknowl-edge the contribution of my co-authors Stephan Winter, Birgit Elias and Monika Sester. Thesupport of the National Mapping Agency of Lower Saxony, Germany (Landesvermessung undGeobasisinformation Niedersachsen) which provided the testing dataset for Hannover (ATKISBasis DLM (www.atkis.de)) is gratefully acknowledged.

59


Salience measures the distinction of a feature from its background. Measures

for automatically quantifying the salience of a landmark in a city environment

have previously been identified and implemented by Elias (2003); Nothegger et al.

(2004); Raubal and Winter (2002), and further extended to include their struc-

tural properties by Klippel and Winter (2005). These measures allow the salience

of spatial features to be quantified and their rank order established.

If we observe the urban structure at a certain granularity, only landmarks of a

certain hierarchical order appear. These are landmarks of similar salience. Global

landmarks have a much larger reference region then local landmarks and their

occurrence in the hierarchy is much less frequent then that of local landmarks.

The size of the reference region therefore allows the establishment of a strat-

ified hierarchical system over several granularity levels. Any location in an urban

space may be characterized by reference to a landmark, i.e. it belongs to a refer-

ence region of a landmark. In the data model presented, the space consists of a

finite number of jointly exhaustive reference regions and their landmarks, across

multiple granularities. The reference regions must be mutually disjoint, as their

generator landmarks are the most salient in the given area. The reference regions

of the landmarks thus constitute a spatial partition. Note that the subset of

landmarks present at a coarser granularity level is present also at the finer level

(e.g., the Eiffel Tower may be considered as a global landmark of Paris, as well as

a local landmark for navigation in its direct vicinity). Thus, landmarks’ reference

regions constitute a hierarchy of partitions, not a hierarchical partition.

Construction of Hierarchies of Landmarks

To build the hierarchy of landmarks, a set of landmarks in the environment has

to be identified. Landmarks at intersections present an important element of

route directions, especially where turns occur (Klippel and Winter, 2005; Michon

and Denis, 2001). If the regional identification method is applied to features

in vista space of each street intersection in an urban space, the collection of

the vista spaces in which the landmarks are evaluated covers the street space

exhaustively. This is due to the high number of landmarks, where the vista space

of each landmark contains several neighbouring landmarks. The identification of

landmarks follows the principles described in Elias (2003, 2006). Landmarks are

selected among buildings at the intersections, based on visibility analysis (Elias,

2006). The result is a list of landmarks for each junction. A more detailed

specification of the methods and data used is discussed in the following section.

The landmarks identified are used as seeds of Voronoi polygons, thus creating

60


a Voronoi partition. Each Voronoi polygon is a minimal convex space in which

the landmark is dominant. Note the use of Voronoi polygons instead of vista

spaces, as vista spaces may overlap.

A leveled hierarchy of landmarks can then be computed by the application

of a recursive selection procedure (Winter et al., to appear 2007)(Algorithm 1).

Algorithm 1: Recursive selection of landmarks through consecutive gran-ularity levels.

Data: L1: Set of landmarks l1k at level 1Result: L1..i: Sets of landmarks at granularity levels iforeach landmark lik ∈ Li do1

compute the most salient landmark lk. max in its immediate2

neighborhood Lik = {lj|dist(lj, lk) ≤ 1, lj ∈ Li}. The distance is defined

as topological distance on Delaunay triangulation;

Put {lk. max|lik ∈ Li} as the set of landmarks at the next higher level of3

salience, Li+1;if the number of landmarks in the set |Li+1| > 1 then4

compute the Voronoi partition and the Delaunay triangulation for all5

li+1k ∈ Li+1;Go to step 1;6

else7

Stop8

All landmarks exist at the lowest, or finest level i = 1. The landmarks present

at higher levels are subsets of the set L1. Each landmark can be characterized

by the maximal level m at which it appears in the leveled hierarchy (lm), and by

the current level i at which it is used in discourse (li), with i ≤ m. The difference

is the reference region associated with the landmark. Landmarks at level 1 can

be considered local landmarks. Landmarks of higher levels (i > 1) are more and

more global in their spatial context.

The Hierarchical Dataset of Landmarks of Hannover

To demonstrate Algorithm 1, consider a testing dataset covering the central part

of the city of Hannover (Figure 4.1). Landmark data were available for 283

intersections, evaluated by the application of the ID3 algorithm to 2200 build-

ings (Elias, 2003, 2006). The landmark identification procedure automatically

identifies building landmarks at each intersection, based on a complex dataset

consisting of the cadastral map, the topographic data set ATKIS, and a high-

resolution 3D city model generated from airborne laser scanning, were available.

Each potential decision point is investigated detecting the salient buildings in its

vista space.

61


Figure 4.1: A map of the test area—center of Hannover, Germany (approximately4 × 4 km, landmarks used are locted in the area highlighted in grey).

Together, 868 building objects were selected as potential landmarks, on av-

erage three for each junction. Due to the short distance between intersections in

urban road networks, many buildings are visible from multiple junctions. There-

fore, the analysis procedure can select individual buildings for more than one

junction as a potential landmark. In the test area there are buildings that are se-

lected only once, but also a building that is selected at 48 junctions as a potential

landmark. Removing the redundant counts, the list of all potential landmarks in

the test area consists of the 295 different building objects.

All objects in this set of potential landmarks are considered equally salient.

For the application of the Algorithm 1, a relative quantification of salience is

necessary. To determine such a graduation of saliency, route-specific aspects may

be considered, or the frequency of selection of a specific building can be assessed.

The second approach was selected for its independence from individual routes

and better relation to the general experience of a city. In this way, visibility is

ranked higher among all characteristics.

The landmark dataset contains landmarks represented as point geometries

and their reference regions as polygon geometries, at the respective hierarchical

levels as identified by Algorithm 1. Together, the hierarchy of landmarks consisted

of 7 levels (Figure 4.2a-4.2d, the starting set of landmarks of Level 0 is not shown).

62


The counts of landmarks retained per level level of selection is shown in Table 4.1.

A textual list of landmarks and a link list of the ancestor landmarks selected in

their reference regions was reconstructed for each level of the hierarchy. This was

derived from the Delaunay triangulation, providing a neighborhood structure of

the landmarks compared.

Table 4.1: Number of landmark objects selected by hierarchical levelLevel 0 1 2 3 4 5 6 7Landmarks 295 95 32 11 4 3 2 1

For further analysis and application of the selection of references for desti-

nation descriptions, only the top six levels of the partition were used. This limits

the number of landmarks and thus simulates a sparser distribution of prominent

features through the study space, which would otherwise not be possible with the

dataset of the size available. A hierarchy of six levels also provides a sufficient

depth of the dataset to demonstrate the functionality of the model.

The hierarchical structure of the dataset is demonstrated on Figure 4.3 for

Levels 3-7. With every finer level of the partition, the number of elements grows

exponentially and thus Level 2 (the least granular level used in the model of

destination descriptions) is not shown. The common names of the landmarks

used are shown in Table A.1 in Appendix 1. The distribution of the landmarks

if Level 2 on the background of the street network of Hannover is shown on

Figure 4.4.

4.4.2 Hierarchies of Districts

Apart from the hierarchy of partitions of the dual regions of landmarks, another

such partition fulfilling our criteria of identifiability of its constituent elements

is the administrative hierarchical partition of a city. Parts of the city, such as

suburbs, referred to by their names, occur frequently as referents in destination

descriptions of locals. Administrative partitions are therefore instances of the

spatial concept of Lynch’s type district, and administrative spatial partitions are a

data structure explored for the automated generation of destination descriptions.

Many researchers have pointed to the fact that administrative hierarchical

partitions of space do not reflect the commonsense conceptualizations, claiming

that people’s conceptualizations of boundaries between parts of the city are usu-

ally not crisp (Alexander, 1988; Dalton, 2006). As hinted by Dalton (2006), the

experience of suburbs of the locals does not exactly match the administrative

partition of the city, but is strongly determined by other structural properties,

63


Level 1

(a)

Level 2

(b)

Level 3

(c)

Level 4

(d)

Level 5

(e)

Level 6

(f)

Level 7

(g)

Figure 4.2: (a): Local landmarks with their reference regions (outlined in black)and neighbored landmarks (black dots). (b)-(f): hierarchical levels of more andmore salient landmarks. (g): the most salient landmark, present through alllevels.

64


Figure 4.3: Representation of the top levels of the hierarchical structure of land-marks (Levels 3 − 7).

Figure 4.4: Spatial distribution of Level 2 landmarks on the background of theirreference regions (black lines) and the street network of Hannover (grey lines).

65


such as the structure of the street network. An example of such structure emerg-

ing from individual experience is the partition of the city by reference regions of

landmarks.

Administrative hierarchical partitions can be represented as tree structures.

The boundaries between the partitions are crisp and well defined (Montello, 2003;

Smith and Varzi, 2000). The boundaries are shared by partitions across multi-

ple levels of granularity, with the coarsest level containing elements of the size

of whole continents or countries. Intermediate levels of granularity are country

specific, and contain elements equivalent to states, other larger regions, counties,

down to cities and suburbs. The partial order establishing the hierarchy is that

of containment. The finest building block of an administrative partition of space

is usually a lot or a house, the smallest region delimited by ownership bound-

aries. The elements at this level of detail are rarely used in communication, and

references to concrete buildings—landmarks are used instead.

In historical cities, the delineation of suburbs is close to the experience of the

locals. This relation between the administrative partition and the experience is

evolutionary and important. Structures of historical cities often emerge by amal-

gamation of villages, usually with a historical market town at the core. Still, in

modern, planned cities, the delineation of the boundaries between administrative

partitions does not necessarily follow the experience of locals. Furthermore, while

administrative partitions such as cities or suburbs have well known names, their

size is too coarse a reference for destination descriptions.

Reference regions of landmarks have a strong experiential basis (i.e., visual

experience of the landmark), and should therefore be considered when construct-

ing experiential hierarchies of districts. Their identifiability is granted by the

name of the generator landmark (“the vicinity of the Eiffel Tower”). The hier-

archical rank of the reference region and the landmark may then be considered

structurally equivalent.

Reference regions of landmarks relate to the empirical phenomenon where a

region around a central landmark embodies the characteristics of a whole suburb

or even a larger region around. Thus, a portion of the city may belong to multiple

regions of the hierarchically higher partition (Figure 4.3).

To include references to districts into destination descriptions , the selection

model must account for a variety of hierarchical structures of districts, such as hi-

erarchical partitions (tree-like structures) and hierarchies of partitions. This must

be done in an integrated manner, as the hierarchical structure may have differ-

ent characteristics at different granularities. Both types of hierarchical structures

occur in the test dataset of Hannover, and are accounted for by the selection

66


model.

4.4.3 Hierarchies of Paths2

In destination descriptions, references to prominent paths (following the defin-

ition of paths of Montello (2005) presented in Section 2.1.2) are frequent. For

pedestrians and car drivers, the network of paths available is represented by the

street network. Consider the directions given to the taxi driver by Stephanie once

again (Section 3). Stephanie assumes that the knowledge of the prominent street

Rathenaustrasse is common, without asking the driver first. She also assumes

that Luisenstrasse is not prominent enough to be localized by the taxi driver

without previous reference to Rathenaustrasse. While the administrative hier-

archy of streets classifies Luisenstrasse and Rathenaustrasse at the same level,

Stephanie assumes that Luisenstrasse may not be known to the taxi driver.

Administrative hierarchies of streets are widely discussed in the literature

(Eppell et al., 2001; Marshall, 2004). Such hierarchies of streets primarily serve

the function of urban traffic and transportation planners. Often the specification

attributing a street to a specific hierarchical level is only vaguely stated and

depends on the designer or traffic planning authority. Administrative hierarchies

of streets are therefore unsuitable for a reliable inference of prominence and the

selection of path references in destination descriptions .

Locals are able to relate information provided by others to their own knowl-

edge. This is only possible to the extent to which the two hierarchies are corre-

sponding. Among locals that do not share mutual knowledge (see Section 2.2.4),

this similarity is necessarily due to the structural properties of the environment,

e.g., the street network and its higher-order functional partition.

The semantic and visual properties of streets are represented by the cumu-

lative semantic and visual properties of the building facades along them. The

semantic and visual properties of a street as such are therefore secondary or in-

herited. The space embodied by the street only acquires these parameters of

prominence once it is experienced by the wayfinder, possibly multiple times to

enforce the strength of the experience. In what follows, an argumentation sup-

porting the consideration of the structural properties of streets in the construction

of experiential hierarchies is presented.

2Parts of this section were previously published in (Tomko et al., to appear 2007). I ac-knowledge the contribution of my co-authors Stephan Winter and Christophe Claramunt. Thesupport of the National Mapping Agency of Lower Saxony, Germany (Landesvermessung undGeobasisinformation Niedersachsen) which provided the testing dataset for Hannover (ATKISBasis DLM (www.atkis.de)) is gratefully acknowledged.

67


Street Network Elements

Experiential hierarchies of elements of street networks cannot be created without

the ability to characterize the prominent parts of the street network by quantifi-

able properties. First, the identification of the elementary part of the hierarchy

has to be performed. As reviewed in Section 2.3.6, the street network can be

modeled with different elements in mind (e.g. street centerline segments, axial

lines, strokes, or named streets).

Street networks are not the result of a partial order established by the part-of

relationship, in a manner similar to districts. No relation of containment or con-

trol can be established between two-dimensional elements of the same kind. A

different relationship thus establishes the partial-order between streets in experi-

ential hierarchies. Note that this is different for different kinds of two-dimensional

elements, e.g. a stroke may consist of multiple named streets, which in turn con-

sist of several street segments.

The model presented in this thesis builds on the selection of a named street

as the building block of the network (Jiang and Claramunt, 2004). A street name

is often the only characteristic of a street that is part of common knowledge.

Thus, named streets allow an integration of the semantic properties of the street

network, which allow to identify a street in a referring expression, such as a

destination description. Further in this thesis, in primal graph representations of

street networks, named streets will be represented as a single edge, in dual graph

representations as a vertex.

Experiential Hierarchies of Streets

The selection of an appropriate measure for the quantification of the importance

of the named street is motivated by its relation to wayfinding behaviour. This

motivation is specific in its cognitive grounding, and complements previous struc-

tural approaches motivated by hierarchical spatial data generalization (Jiang and

Claramunt, 2003).

Frequently experienced parts of the network are prominent, and thus rank

high in the hierarchical mental representations. Street connectivity influences

the pattern of urban movement flows and determines the intensity of learning

the urban layout. The prominence of streets in the network thus relates to the

structural properties of the street network.

Structurally prominent parts of the street network are not only frequently

experienced directly by wayfinders, but this experience is further strengthened

68


by secondary influences. Indirect sources, such as Web pages and news articles

reflect on the prominent streets, further supporting the individual experience.

Centrality Values in Street Networks

The inference of the shared experience of parts of the street network is related

to the structural properties of the network. The hierarchy created by experience

needs to take into account the likelihood of the usage of a specific street, not only

its central aspect. A comparison of the values of centrality measures and their

distribution, starting with regular street networks, follows. By the introduction

of irregularities in the regular patterns, the distribution of the centrality values

is altered and the effect is evaluated from the perspective on the impact on the

shared experience of individual streets. This allows centrality measures to be

identified, that reveal structural properties of named streets contributing to their

experiential prominence.

Rectangular grid patterns consist of perpendicular streets forming blocks.

All junctions have the same degree and are thus identical in their local structure.

Such urban layouts are typical for modern planned cities. Downtown areas of

major US and Australian cities follow this pattern, as well as some European

planned cities, e.g. Barcelona. Some of these cities have a few streets intersecting

the grid pattern diagonally. Such streets are usually well known.

Betweenness centrality reveals the relative importance of these streets, as

shown by comparison with the degree and closeness centrality. Figure 4.5 presents

a grid pattern and its dual graph representation consisting of 6 orthogonal streets,

forming a grid of 2 × 2 blocks. The dual graph analysis was performed using

the software Pajek (Batagelj and Mrvar, 2006). The graphs reveal the bipartite

structure of the north-south and east-west streets. The betweenness values of

the network of named streets are all equal. If we consider named streets as the

building element of the grid-like street network, no structural difference between

the individual streets is revealed by in any of the degree, closeness or between-

ness centrality measures. If the element of analysis is a street segment, higher

betweenness and closeness centrality values are attributed to the central part of

the grid. As all the streets are intersected by the same number of connecting

streets, degree centrality remains uniform in the whole grid.

In such a regular grid, however, the hierarchical ranking of streets by be-

tweenness and closeness is identical. Hence, irregularities in urban layouts that

impact on the perception of the city as such also cause variance between the cen-

trality values. The addition of a diagonal street (Node 7 in Figure 4.6b) in the

69


(a) Street network (primal graph rep-resentation)

[0.10] 1

[0.10] 2 [0.10] 3

[0.10] 4

[0.10] 5[0.10] 6

Pajek

(b) Street network (dual graph representation).Labels represent betweenness values of streets.

Figure 4.5: Graph representations of a grid network of named streets.

grid network leads to a change of betweenness values of the streets. The north-

south streets intersect all of the streets in the network and thus would lie on most

shortest paths. They rank on top of the hierarchy when ordered by betweenness.

The newly introduced shortcut follows in the ranking. The ranking by closeness

centrality or degree centrality would, however, not reflect the direct experience

of the urban structure appropriately.

In a star-like network (Figure 4.7), the insertion of a shortcut (Street 4)

changes the reachability of the peripheries involved. Values of betweenness cen-

trality reflect this change in a manner reflecting the likelihood of common ex-

perience of a network element. Betweenness centrality thus reflects better the

evolution of experiential hierarchies of streets of wayfinders.

It is the occurrence of shortcuts between internally highly connected sub-

graphs that motivates the use of betweenness for the reconstruction of the hier-

archical structure of the street network. In a street network, such subgraphs may

stand for districts, where the internal connectivity of the street sub-network is

higher than that in the remainder of the city.

The dual graph representation of the street network from Figure 4.7 reveals

the structural changes caused by the shortcut (Figure 4.8). By the addition of

Street 4, the two distant peripheral parts of the network become directly con-

nected. A new urban core is created by the triangle 1, 3 and 4. The individual

importance of Streets 1 and 3 decreased with the introduction of the Street 4.

The measure of betweenness reveals the alteration of experiential prominence of

70


(a) Street network (primal graphrepresentation)

[0.08] 1

[0.05] 2

[0.05] 3

[0.05] 4

[0.08] 5

[0.08] 6

[0.07] 7

Pajek

(b) Street network (dual graph representa-tion). Labels represent betweenness valuesof streets.

Figure 4.6: Graph representations of named streets grid with a diagonal street—shortcut.

Figure 4.7: A primal graph representation of a star-shaped street network withstreets labeled 1, 2, 3 and 4. Street 4 forms a shortcut.

streets in the network. It preserves the high prominence of Street 2 and reflects

the lowering of the importance of Streets 1 and 3. Before the insertion of Street

4, betweenness values for the Streets 1, 2 and 3 in the network were equal, 0.44.

After the insertion of Street 4, Street 1 has a betweenness centrality value of 0.41,

Streets 1 and 3 have values of 0.23 and Street 4 a value of 0.18.

Closeness centrality fails to reveal the alteration of this street network ap-

propriately. The closeness centrality value of Street 2 decreases as Street 4 is

inserted. Street 2, however, remains the only means of access to a significant

portion of the graph. Also, the relatively high values of closeness of the streets on

the peripheries of the graph compared to the central streets have little justifica-

tion from the experiential point of view. Thus, closeness centrality fails to reveal

the relative importance of the streets to the overall structure of the city. The

importance of Streets 2 and 4 would become even more prominent if a partition

71


(a) Dual graph representation of the street net-work, with betweenness centrality values.

(b) Dual graph representation of the street net-work, with closeness centrality values.

Figure 4.8: Dual graph representations of streets in a star-shaped street networkwith added shortcut (Street 4).

of the network into a suburb was introduced. The peripheral, cohesive parts of

the graphs can be clustered into a district. In such a case, the significance of

Streets 2 and 4 as links in the functional structure of the city would be further

emphasized.

Due to its local character, degree centrality does not provide a measure of

prominence of a network element outside of its immediate neighborhood. In natu-

rally evolved spatial transport networks with high degree of asymmetry, closeness

centrality does not provide a reliable measure of hierarchical importance of a net-

work element in the overall network. It distorts the hierarchy by assigning higher

values to the streets in the core of the network. Side-lanes and alley-ways located

at the geographic center of the area of interest will always get high closeness

values, as long as they form loops (cycles) and thus do not lie on the periphery

of the graph. This structural property, however, does not necessarily make them

prominent.

Betweenness centrality reflects the probability of a street to be selected by

a frequent wayfinder for a trip within the street network. With the increasing

number of trips performed, the likelihood that betweenness approximates the

agent’s experience of the urban environment increases.

Experiential Street Hierarchies and the District Partition

The structural role of the streets facilitating movement between the functional

partitions of the city into suburbs should be considered for a refined ranking of

streets in experiential hierarchies.

72


The importance of a street is relatively higher if it provides the only access

to a district, for instance a suburb. It is, however, more common to have several

streets that connect the same suburbs. In turn, their respective betweenness val-

ues are lower, as the streets are part of alternative access routes. This is frequently

the case in modern agglomerations with regular grid patterns in the center of the

city. Those streets, however, are still prominent, as their prominence is in turn

supplemented by their membership in a structurally important suburb. As illus-

trated by Figure. 4.9, the two streets present alternatives for travel between the

two nodes. Their respective betweenness centrality values would, in a network,

be equal to half of the value of a single street connecting the two nodes. These

streets, however, connect a central suburb to to peripheral suburbs, which may

change the perception of their prominence in a city.

Figure 4.9: Alternative streets with equal betweenness related to their suburbcontext (primal graph representation).

The higher-order structural embedding of a street contributes to its hierar-

chical ranking in experiential hierarchies. The urban district partition thus rep-

resents an overlaying functional structure over the basic structure of the street

network. To fine-tune the experiential hierarchy of streets, considering the struc-

tural relations between the districts should be explored.

In order to combine the functional and structural characteristics of the urban

structure, the betweenness centrality of districts is considered as a second para-

meter influencing the rank of streets in the network. A second graph is derived

from the partition of districts where suburbs are the nodes of the graph, edges

adjacency relationships between districts, as facilitated by streets. Betweenness

centrality allows for consideration of the following structural properties in the

street-district relationship as:

• Districts of high betweenness are crossed by a high proportion of possible

trips in the network.

• Streets contained in districts of high betweenness are likely to be experi-

enced more often.

Based on the betweenness centrality values of streets and districts, a novel

measure for ranking of the streets in an experiential hierarchy of the street network

can be introduced.

73


More formally, these notions are denoted as follows:

• Let i denote a street in the street network;

• D the set of districts in the city;

• Di the set of districts intersected by the street i;

• dij a district out of Di;

• CBi the betweenness centrality value of the street i;

• CBdi

j

the betweenness centrality of the district dij;

• ndkdlthe number of shortest paths linking two districts dk and dl of D;

• ndkdl(di

j) the number of shortest paths linking two districts dk and dl that

contains dij.

Betweenness centrality of a district CBdi

j

then equals:

CBi =

∑

dij 6=dk 6=dl

ndkdl(di

j)

ndkdl

(4.1)

The experiential rank value Ei for a named street i in the experiential hierarchy

of the street network based on the betweenness centralities of the street network

(Eq. 2.3) and the district partition of space (Eq. 4.1) is then defined as follows:

Ei =∑

j

CBi × CB

dij

(4.2)

The value of Ei is calculated based on the adjacency matrix of districts and

streets. This matrix contains relations of districts intersected by streets in the

street network. If such a relation exists, the betweenness centrality CBi value of

the street i is multiplied by the betweenness centrality values of the districts in

Di (CBdi

j

), and the resulting values are summarized. Ei is not a normalized value,

and thus can be greater than 1. The values of Ei are calculated only for the

purpose of ranking, their direct comparisons between different street networks

are meaningless. Thus, Ei is an ordinal measure.

The results of a structural analysis of the street network and its higher-order

functional partition provide a relative estimate of the prominence of a street

in the city structure, presented as a ranking of the streets in an experiential

hierarchy. Such ranking allows the abstraction of the urban network at different

granularities, preserving the inherent logic of its structure. For the application

74


and verification of the method on a dataset of the street network and district

partition of Melbourne, Australia, see Tomko et al. (to appear 2007).

The Hierarchical Dataset of the Street Network of Hannover

To demonstrate the distribution of the values of Ei, the dataset of the the street

network of Hannover and the second level partition of the city by landmark

reference regions (Figure 4.2b) was used. The second level partition was selected

as it presents the basic partition of the city into reference regions of all the

landmarks considered. The consideration of partitions of coarser granularities

would lead to a loss of detail, as more streets would connect a single district.

First, the street centerline dataset was processed in order to merge streets

segments into street names. The unique Street ID attribute was used to merge

related street segments. Note that this was considered to be a more reliable

method then merging by real street names, as the street name attributes were

incomplete.

The betweenness centrality analysis was performed in the space syntax soft-

ware Mindwalk (Figueiredo, 2002) on a street network of 1350 named streets, out

of which 394 were located in the study area of the center of Hannover. Mindwalk

implements a computationally efficient version of betweenness centrality, called

fast choice. It considers only one random shortest path between each pair of

nodes in a graph, instead of generating all the alternative shortest paths. In

larger networks, such as the one used in the example, the differences in central-

ity values resulting from fast choice are statistically insignificant. This can be

simply verified by multiple analysis of the same network, noting that the values

computed do not change. The result of the analysis is a vector of betweenness

centrality values CBi for all the named streets in the study area.

In parallel, a matrix recording the connectivity of districts by named streets

A394×32 was created (the 394 streets in the study area intersect 32 Level 2 dis-

tricts). The matrix was based on a link-list of named streets and districts con-

nected through them. The reduction from such a 2-mode network into a 1-mode

network was performed in the network analysis software Pajek (Batagelj and

Mrvar, 2006). Consecutively, the vector of betweenness centrality values for the

districts CBdi

j

was calculated. The resulting values are reported in Table 4.2 and

the connectivity of the districts is shown in Figure 4.10.

Note that the adjacency graph of the districts (Figure 4.10) is different to

the Delaunay triangulation of the landmarks. Named streets connect multiple

districts which are thus considered connected, although they may not be adjacent

75


Table 4.2: Betweenness centrality vector CBdi

j

of Level 2 districts of central Han-nover

District CBdi

j

District CBdi

j

2_H074YH2 0.333685403 cont.2_H03WTT1 0.327886062 2_H05TWH4 0.023800516

2_H03PNBO 0.101352828 2_H05JJHO 0.023800516

2_H04SBR1 0.09743539 2_H06KOBS 0.023800516

2_H063YJC 0.071779952 2_H03PO3Z 0.022947207

2_H01P2HN 0.069681854 2_H09DX6M 0.016543923

2_H05V43Q 0.063629053 2_H04PTS0 0.015997843

2_H05V4AW 0.057025478 2_H06E9I7 0.005319844

2_H03P85A 0.055171356 2_H05T3WR 0.004174156

2_H03Q6S0 0.052112675 2_H03NG5F 0.003225806

2_H01F6M0 0.044038213 2_H05MF0G 0.002133835

2_H03WUC7 0.038127862 2_H06Y0NB 0.001971326

2_H01O23Z 0.034489386 2_H01FM8E 0.001171779

2_H01BHXG 0.034489386 2_H05RN6G 0

2_H097TLK 0.029566636 2_H05IWO5 0

2_H03NGBE 0.025948269 2_H05V49E 0

2_H03PO1O 0.025144542

Figure 4.10: The graph of the connections of the level 2 districts by named streets(betweenness centrality values in square brackets).

76


in the Voronoi partition of the space.

The vectors CBi and CB

dij

, and the adjacency matrix of named paths and

districts A were combined to calculate the experiential rank values Ei for each of

the named streets, according to Equation 4.2. The plot of the Ei values for all

the named streets in the study area, in descending order, is shown in Figure 4.11.

Figure 4.11: Distribution of experiential rank values of the streets of Hannover.Left of the dotted line are the streets of experiential rank above mean.

The values of Ei range between 0 and 0.147. The mean value calculated from

the sample was x = 0.0019 with a standard deviation in the dataset σ = 0.0113.

The distribution of experiential rank values follows a power-law distribution

(Newman, 2005). The dotted line splits the named streets with Ei values above

the mean value (left of the line) from the named streets with Ei values under the

mean value (right of the line).

There are in total 33 streets with Ei above the mean x, and only 9 such that

their experiential rank value is more than a σ above x (Table 4.3).

Individual experiential hierarchies are continuous rankings, and it is impos-

sible to draw a line separating prominent and non-prominent streets. It is, how-

ever, possible to approximate this limit by the mean value in the distribution.

The streets of above-mean experiential rank values will be called prominent. The

bulk of the streets in the hierarchy are below the mean value of prominence. In

the case of the city of Hannover, the streets around the mean value of Ei are

more than a thousand times as prominent as those with the lowest values (null

values were not considered).

Figure 4.12 shows the 33 most prominent streets of the study are in central

77


Table 4.3: Most prominent streets of central Hannover

Rank Street ID Street name Rank Street ID Street name1 N01FU95 Engelbosteler 18 N20B4C7 Celler Straße

Damm2 N00E2TA Bremer 19 N01FUV6 Kurt-Schumacher

Damm Straße3 N20C8LF Nienburger 20 N01FUFK Herrenhauser

Straße Straße4 N202568 Konigsworther 21 N01FUKU Herrenhauser

Platz Allee5 N20256S Leibnizufer 22 N01FVRO Marienstraße6 N20256D Bruhlstraße 23 N01FUVH Maschstraße7 N01FUQO Joegerstraße 24 N01H2FY Weddigenufer8 N01FUSL Friederikenplatz 25 N20C8N2 Lodyweg9 N01FVOF Osterstraße 26 N01FVVY Burgstraße10 N202566 Goethestraße 27 N2025VG Schmiedestraße11 N01FVPX Herschelstraße 28 N202578 Platz der Gottinger

Sieben12 N20GR7U Culemannstraße 29 N01FUTR Goseriede13 N20GR5B Walter-Großmann 30 N01FUJS Karmarschstraße

Weg14 N20256U Lavesallee 31 N01FVP7 Andreastraße15 N202569 Calenberger 32 N01FUU3 Leinstraße

Straße16 N01FVYX Arndtstraße 33 N01H2TY Große

Packhofstraße17 N20DDIH Friedrichswall

78


Hannover in an overlay with the boundaries of the study area and the second-level

Voronoi partition of the area by landmark reference regions.

N20C

8LF

N20G

R5B

N01FU

95

N01FU

KU

N00E2TA

N01FVRO

N01FUFK

N20

B4C7

N20

256U

N01FV

PX

N01FUQO

N01F

VO

F

N202569

N20GR7U

N20256S

N202566

N20DDIH

N20

2568

N01

FU

JS

N01FUV6

N2025V

G

N01H2FY

N01FU

VH

N01FUU3

N01FVYX

N01FV

VY

N20C

8N

2

N01FUSL

N20256D

Figure 4.12: 33 most prominent streets of central Hannover (named streets ofEi ≥ x), labeled by street ID.

The joint consideration of the street network structure together with the sub-

urb partition of the city allow for a reliable identification of the most prominent

streets of the street network, as shown on the example of central Hannover and

in a previous study of a larger area in Melbourne, Australia (Tomko et al., to

appear 2007). The result illustrates the plausibility with which the novel measure

reveals the experiential hierarchy in an urban network.

The prominent streets identified represent empirically important connectors

of central Hannover, including the prominent streets along the boundary and

across the study area: Engelbosteler Damm, Bremer Damm, Nienburger Straße,

Friederikenplatz, Bruhlstraße and Leibnizufer, among others.

The exploration of the spatial distribution of the prominent streets shows

that they are relatively evenly distributed across the study area, without mani-

festation of the boundary effect. This was expected, as the constituting measure,

betweenness centrality, is not susceptible to boundary effects. The distribution

corresponds with the empirical experience where a relatively small proportion of

79


streets in a city form a cognitively important skeleton structure. This means that

in any urban structure there is a relatively large number of streets with a low

prominence, and only a few prominent ones.

The empirical success of route direction communication in our daily lives

suggests that there is a large overlap in the structures of our spatial knowledge,

and that the knowledge of the prominent parts is common. Thus, a certain de-

gree of difference in individual experiential hierarchies is not a barrier to reaching

understanding. The match between the prominent parts of the hierarchies of the

speaker and of the hearer is high when most of their elements are identical, but

not necessarily ranked in the same order. Due to the power-law distribution of

street prominence values in the experiential hierarchies, highly prominent streets

have values higher by magnitudes than those of low prominence. Thus, while the

individual rankings in the mental representations of streets of a speaker and a

hearer may be different, the sets of the prominent streets will by largely overlap-

ping. This also allows the use of the proposed objectivized experiential hierarchy,

for an automated construction of route directions by an automated service. Even

a small sub-set of all the streets in the street network may provide a sufficient set

of potential referents for place and route descriptions in the area in consideration

(Kuipers et al., 2003).

4.5 Integration of Elements of the City in an

Experiential Hierarchy

As shown, it is possible to construct cognitively motivated, experiential hier-

archies of diverse elements of the city. The next step is the construction of a

single, integrated experiential hierarchy of the urban environment, necessary for

a coherent selection of references in destination descriptions. The interdepen-

dent insertion of references to elements of different types relies on such integrated

structures. The development of an integrated hierarchical dataset of multiple

elements of the city, allowing for the selection of the most relevant referents for

destination descriptions, is critical.

The three types of elements of the city are organized in hierarchies of very

different properties:

• named streets are organized in a rank-order which is the function of the

frequency of occurrence of their experiential rank values in the dataset;

• landmarks may be ordered in a rank of similar properties, but are also linked

80

M. TOMKO 4.5. INTEGRATED EXPERIENTIAL HIERARCHY

through the properties of their reference regions in a containment hierarchy

of partition, which is an m : n relationship;

• administrative districts may be organized in a containment hierarchy, mostly

by their size. However, through the property of containment, they also

structurally integrate paths and landmarks.

Other relationships are also possible. The structure of the street network may

determine the experiencing of the districts, or even their genesis. Dalton (2006)

for example suggests a redefinition of suburbs as a function of the structure of

the street network.

Of course, these types of elements of the city form have also relations at the

same level, not only across hierarchical levels or granularities. Paths connect dis-

tricts, while landmarks have a perceptual influence on their reference regions and

thus give context to districts (i.e., as seeds of the Voronoi partition). Landmarks

are also experienced by wayfinders navigating along paths, they are en route.

Figure 4.13 schematically depicts these possible relationships.

Figure 4.13: Schema of relations between heterogeneous types of elements of thecity in integrated hierarchies.

As shown in Figure 4.13, the integration of the individual hierarchies is possi-

ble through the hierarchical structure of districts. They provide a dual structure

for the organization of landmarks, and also provide a containment hierarchy for

paths. Thus, the structural embedding of the ranked order of paths and the hi-

erarchy of landmarks in the structure of districts will be used in the model of

destination descriptions proposed in Chapter 6.

To distinguish between a referent to districts and landmarks, the adminis-

trative names for the suburbs of Hannover will be used at the coarse levels of

granularity in the integrated structure of landmarks and districts, as the tracing

of the limit between well-known and presumably not-known districts is still the

subject of future research.

81


4.6 Concept of Distance in Hierarchical Struc-

tures

The concept of topological distance is operationalized upon the hierarchical ur-

ban structure proposed in order to execute the model for selection of references

for destination descriptions. The topological distance of two elements in the hi-

erarchical model of the urban structure is computed as the minimum number of

elements of the same type and granularity, lying between the two input elements.

Thus, a pair of neighboring districts of the same granularity has a topological

distance 0. Imagine two districts of the same granularity dA and dB, separated

by a third district of equal granularity dC , such that dA and dC , and dB and

dC are neighbours. Then, the topological distance |dAdB| = 1. Note that if

two elements have different granularities, topological distance is not defined. If

the spatial model is analysed at coarse granularity, finer-granularity elements are

not present. In such case, the topological distance is calculated as the minimal

topological distance of the coarser of the two elements in question, and a super-

ordinate (or ancestor) element of the finer element. The notation Supere will be

used for the set of elements that are superordinate to e, and the notation Supe

will be used for a parent, i.e. directly superordinate, element of e.

To calculate the topological distance in the hierarchical spatial data struc-

ture, topological relations of neighbourhood between the elements are analysed.

Two elements are considered neighbouring if they share a boundary and are con-

nected. In the spatial data model used, two districts are neighbours if they share

a boundary and are connected by a path. Landmarks are neighbours if their refer-

ence regions are neighbours and they are connected by a path. Paths are modelled

as a concatenation of finest-granularity districts through which a named-street

leads. Thus, two paths are considered neighbours if they share one or more of

their defining districts. By analogy, the relations of neighbourhood between a

district and a landmark are defined. Finally, a path is considered neighbour of a

district or a landmark if the district or the reference region of the landmark are

among the defining districts of the path.

82

Chapter 5

A Generic Model of Destination

Descriptions

The cognitive environment of the speaker and the hearer, i.e. the context in

which the communication of destination descriptions occurs, determines the se-

lection of references. Modeling context is, however, a non-deterministic problem

(Dey, 2000). The parameters of context relevant in a specific task depend on the

individual. As noted by Clark and Marshall (1981, cited in Sperber and Wilson

(1982)), physical co-presence, linguistic co-presence and community membership

are factors facilitating the inference of common knowledge among communica-

tors. Each of these broad groups of parameters could be further decomposed in

detailed parameters of context. A system developer will invariably make subjec-

tive assumptions about the users of the system developed. The more assumptions

a developer commits to, the less general and adaptive the resulting system will

be. The approach presented in this thesis therefore relies on a minimal set of

assumptions about the cognitive environment of the speaker and the hearer.

5.1 Context Specifications for Modelling Desti-

nation Descriptions

The model of destination descriptions in urban environment presented is based

on a minimal specification of the context in which the destination descriptions di-

rections are communicated. These specifications are focused on the characteristic

of the hearer:

A-priori spatial knowledge (Super condition) The hearer is assumed to have

common spatial knowledge formed by experiencing the space during navi-

gation in a finite number of previous trips. The extent of this knowledge is

83

CHAPTER 5. A GENERIC MODEL OF DESTINATION DESCRIPTIONS

not made explicitly known to the speaker, and it is therefore left unspecified

in the model proposed.

Functional perspective A functional perspective on the urban structure is de-

termined by the selection of the means of transport, ostensively disclosed

to the speaker.

Co-presence The references retrieved will be relevant from the perspective of

the current location, physically or virtually shared by the hearer with the

speaker at the moment of selection of a reference by the speaker.

Community membership is reinforced by the requirement of the hearer to

have at least coarse spatial knowledge of the environment, i.e. the hearer may be

considered a local. The knowledge of the hearer will be acquired by perceiving

the environment while navigating in the city.

The requirement of a functional perspective on the spatial knowledge of the

hearer links to the condition of possession of a-priori spatial knowledge. It as-

sumes that the a-priori spatial knowledge of the hearer is conventional in nature,

i.e. the means of transport used to follow the directions provided to the hearer

allows the use of the spatial knowledge of the speaker. This requirement allows

to classify the elements of the city to the five types of Lynch by their function.

For example, the streets in the street network accessible by car will be used as

paths by a taxi-driver, and the canals of Venice will be used as a network of paths

in the case of a Venetian gondolier.

Co-presence allows the speaker to assume the spatial context the hearer will

have on the urban environment when interpreting a given reference. This co-

presence need not be physical, but can be virtual (Zhao, 2003), or projected

(Gerrig et al., 2001). Projected co-presence allows the speaker to infer the spatial

context of the hearer at the moment of interpretation of a reference provided in

the destination descriptions. A reference is interpreted by a hearer in the spatial

context specified by the previous reference, or in case of the first reference of

the destination descriptions in the context of the start of the route. Co-presence

requirement has a fundamental impact on the selection of references in destination

descriptions.

These specifications of the hearer’s context reflect the three major groups of

context characteristics. They represent the minimal assumptions made about the

hearer by the speaker, here modeled as a computational system.

84

M. TOMKO 5.2. MODEL CONSTRAINTS

5.2 Model Constraints

As presented in Section 3.4 and shown in Figure 3.4, the common knowledge of

the speaker and the hearer C, is the intersection of the mental representations of

the reality by the speaker (S) and the hearer (H), acquired by perception of the

urban environment. The references made by the speaker in a given environment

can only be to elements e1..m of the set S, e1..m ∈ S. To reach understanding,

however, these references must also be members of C. Otherwise, the hearer

will not be able to identify the referents in her or his spatial representation and

understanding will not be reached. In the model presented, a necessary and

explicit assumption is made: the understanding of the speaker and the hearer is

possible, i.e. the relative ranking of elements in the spatial mental representations

of the hearer and the speaker is preserved {ea, eb| if eSa > eS

b then eHa ≥ eH

b } (see

Section 3.4 for the discussion on reference selection).

The constraints of the model presented are summarized as follows:

• Application exclusive to urban (anthropomorphic) environment. In natural

environments the conceptual elements constituting the structure of space

may be different (e.g., mountains and rock formations acting as landmarks).

Their salience and ranking is not modelled, and the principles of their se-

lection may differ.

• Extent of distortions in spatial knowledge of the hearer allowing understand-

ing of the speaker. The preservation of the relative ranking of the elements

in the hierarchies of the sets S (operationalized as an experiential hierarchi-

cal data set) and H is assumed. The knowledge of other aspects of context

of the hearer, such as community membership, may be used to improve the

inference of spatial knowledge of the hearer and the consequent selection of

references. This is, however, not considered in the model proposed. The

combination of the inferential model of destination descriptions with agent

based approaches to assess individual spatial knowledge is envisaged (see

Chapter 7).

5.3 Structure of Destination Descriptions

In destination descriptions, the route is not described by the speaker to the hearer

in full detail. Only the referents selected by the speaker during route planning as

the most relevant are communicated to the hearer. Note that the route imagined

by the speaker fulfils the speaker’s own route planning criteria.

85


Imagine the speaker, mentally travelling through the mental representation

of the route and its surrounding vista spaces, and referring to prominent entities

which are part of this mental representation. The selection of the type of referent

(i.e., path, landmark or district) is determined by such factors as the prominence

of all potential referents and the structure of the route described. The referent

selected should be the most relevant one in the given context.

The immediate, visually accessible, vicinity of the route is called route con-

text. In this model, the route context is considered in the process of selection of

referents for the destination description. The notation routes,t is introduced for

a given route retrieved by the speaker, either human or a computational system.

The route is modelled as a concatenation of paths or their parts. A route is

considered equivalent to the route context, as it is composed of districts of finest

granularity available in the experiential hierarchical spatial dataset. The first

district of the set representing the route is the start s, where the speaker and

the hearer are co-located at the moment of selection of the references. The last

district of the routes,t is the destination t.

Destination descriptions can therefore be represented as a serialization of

references of increasingly fine granularity, selected from the set C:

r1, r2, . . . , rn

This can be illustrated on the example of the Castle from Section 3.4 (Fig-

ure 3.5).

theCastle, theNew Palace

e5, e11

The last reference rn may be identical to the reference to the destination

itself, t, and will be of the finest granularity in the destination description. On

the other hand, the first reference r1 will always be of the coarsest granularity

in the destination description. This reference can also be called the initial ref-

erence. Every consecutive reference is then of finer granularity (and thus lesser

prominence).

Thus, any referent selected must provide the most relevant information in

the given context, in order to construct a referring expression. This context

is, however, altered by the utterance of the previous reference. The process of

selection of any consecutive reference is then equivalent to the process of selection

of the first reference included in the destination description. This suggests that

the process can be modelled as a recursive selection of the most relevant reference

86

M. TOMKO 5.4. RELEVANCE OF A REFERENCE

in the context specified by the previous reference. Note that it is assumed that

the start of the route is explicitly known by both the speaker and the hearer.

5.4 Relevance of a Reference

The application of the principle of relevance (Definition 3, page 23) to the se-

lection of references for destination descriptions requires a cognitively plausible

operationalization of cognitive effort and cognitive effect in a given context. This

will allow to reference the element of highest relevance in the candidate set. The

quantification can hardly be absolute. A relative comparison of the estimates of

the cognitive effort needed to interpret the potential referents is therefore chosen.

The first element that satisfies a set of rules assessing the relevance of a reference

will be selected. This approach is commonly called lazy evaluation.

In the model proposed, the relevance of a reference r to an element, in a given

context, is operationalized as a function of the prominence of r, and of the distance

of the element represented by r from the start s (Equation 5.1), in a model of

an environment. The routes,t, or more specifically, its start s and destination t

provide the parameters of context required by the principle of relevance. The

distance of the element, as well as the prominence of r are evaluated relative to

the distance from s and the prominence of t.

relevance(s,t)r =

(

rankr

distancer

)(s,t)

(5.1)

The more prominent an element of the environment is, the less effort is

required from the hearer to relate the reference made by the speaker to her or

his mental representation of the element. Furthermore, no references of lower

prominence than that of t are relevant, as they would increase the cognitive effort

of processing them, and would not provide any cognitive effect to the hearer. On

the other hand, distance from the referent increases the hearer’s cognitive effort,

as the ambiguity of the reference increases. The greater the distance between the

current location of the hearer (s) and the element represented by the reference r,

the larger is the choice set of elements that has to be mentally searched through.

Distance is thus a measure enabling the cognitive effort required to process a

reference to an element to be estimated.

The reference selected for inclusion in destination descriptions must balance

the requirement to provide a reference to the most prominent reference possi-

ble, with the requirement of referring to a reference close to the current spatial

context s. Not only the balanced consideration of the two factors allows for the

87


evaluation of the relative relevance of a reference, it also allows for the avoidance

of trivial references. Trivial references are references requiring low cognitive ef-

fort to process, but which provide low cognitive effect in the given context. The

inclusion of trivial references results in an excessive number of references in the

resulting destination description. Such a destination description does not satisfy

the requirement of a referring expression, due to redundant references.

To preserve cognitive plausibility, the model presented does not require the

use of Euclidean distance of the potential referents in a given context. Instead,

topological distance between elements in the hierarchical structure of the environ-

ment is considered.

5.5 Rules for Selecting District References

To select the most relevant reference r in a given spatial context, the topological

relation of the spatial context defined by the start of the route s, its destination

t, and the distance from the potential references is evaluated in the hierarchical

structure of the city.

Consider a dataset CD containing the set of districts covering the route

routes,t from start s to destination t. The dataset is organized as a hierarchical

partition of space. The identification of references to be included in destination

description in the context of s in this hierarchical dataset can be codified in a set

of rules.

First of all, two specific scenarios have to be distinguished in which destina-

tion descriptions can not be provided. First, the start s and the target t must be

part of the spatial knowledge of to the speaker, or in the case of a computational

system, included in the database CD. Second, the route described must have a

distinct start and destination. As a consequence, the start s and the destina-

tion t may not be specified as identical. The Rules 1 and 2 verify these basic

requirements for the provision of destination descriptions.

1. Start s and destination t are members of the set of elements CD (s, t ∈ CD).

2. Start and destination must not be identical (s 6= t).

Consecutively, if s and t are neighbours, or their directly superordinate elements

are neighbours, the topological distance between the specification of the start

and the specification of the destination is insufficient to generate destination

description. At the granularity at which the destination description of t was

88

M. TOMKO 5.5. RULES FOR SELECTING DISTRICT REFERENCES

requested it is not possible to construct a referring expression specifying the

destination with more detail. Additional references, providing more detail than

the direct reference to t, are not available in the set CD. In the case of the

communication of two people, this situation occurs when the common spatial

knowledge is shared only at coarse granularity. While it is possible that this

is a sufficient specification of the destination for certain hearers, in a general

case turn-based directions from the last granular reference to the destination are

provided. This requirement can be formalized as the Rules 3 and 4:

3. The start and the destination should not be neighbors.

4. The start and the destination should not have neighboring direct superor-

dinate elements.

The higher the prominence of a referent, the higher is the cognitive effect

and lower the cognitive effort of processing the reference. As the reference r is

necessarily a superordinate (or ancestor) elements of t, the superordinate elements

of t are evaluated. Sets Supert of superordinate elements of t, and Supers of

superordinate elements of s can be extracted from the hierarchical dataset CD.

The set Supert is then the candidate set for the reference r:r ∈ Supert.

The inclusion of the reference r in the context of s must provide relevant in-

formation to the resulting destination description. Thus, the topological distance

of the start s to the reference r should be low, but not trivial. In the hierarchi-

cal partition of the environment CD, sets Supert and Supers have necessarily at

least one common element—the root of the hierarchy. If an element is common

to Supert and Supers, a reference to this element does not provide any informa-

tion value in to the hearer in the spatial context of s. Such a reference has no

pragmatic information content and is thus trivial. If the condition is not fulfilled,

an element of finer granularity should be evaluated.

Finally, the Rule 6 assures a minimal topological distance of 1 between the

referent r and the start s (dists,r = 1), in order to avoid references of low relevance.

5. Element r must not be shared by Supers and Supert, (r /∈ Supers).

The cognitive effort to process such a reference is not balanced by the cognitive

effect (i.e., information value) it provides. If the condition is not fulfilled, an

element of finer granularity, satisfying all conditions above should be selected.

This rule is derived from a strict interpretation of the principle of relevance.

6. Element r should not be neighbor with an element in Supers.

89


The Rules 1-6 can be formalized as function ref and represented by the

Algorithm 2. Note that in Algorithm 2, Rules 5 and 6 are applied in one step

(line 11-13).

Algorithm 2: ref – selection of the reference r in the context of s, t.

Data:CD: Dataset–hierarchical partition of districts;s: current location (start district)t: destination district.Result: The reference r for destination description to t in the context of sConstruct candidate sets of superordinate elements of s and t, ordered by1

decreasing granularity (Supert = [dn, . . . t]) and Supers = [dn, . . . s]);Compare Supers, Supert;2

case (s, t /∈ CD)3

Error: cannot generate destination description, stop4

case (s = t)5


case (s, t = neighbors)7

Route too short for destination descriptions, switch to turn-based8

directionscase (Sups, Supt = neighbors)9

Route too short for destination descriptions, return t and switch to10

turn-based directionsforeach dx ∈ Supert) do11

if (dx /∈ Supers) ∨ (∀dy ∈ Supers, dx ∩ dy = 0) then12

put dx in list;13

order list in order of descending granularity;14

return r = first d from list;15

In order to demonstrate the rules specified, consider a hierarchical partition

of the environment into districts (Figures 5.1 and 5.2). Imagine a route routes,t

through this environment. This route is defined as a series of the finest-granularity

districts of the environment. Consider the start of the routes,t s represented by the

district d122, and the destination t by the district d322 (routes,t = routed122,d322).

The application of the rules consists of the assessment of the information

value provided by the potential referents from the sequence of superordinate ele-

ments of the destination t = d322 (including the destination itself):

d0, d3, d32, d322

Rules 1 to 4 are satisfied by the composition of the dataset and the structure

of the route itself, and a destination description can therefore be created. Rules 5

and 6 are then used to select the first reference of the destination description of

90


(a) Schematic representation of space partitionedinto districts CD.

(b) Schematic representation of the routeroutes,t.

Figure 5.1: Schematic representation of (a) the hierarchical partition of space and(b) the schema of a route through this partition.

Figure 5.2: The hierarchical representation of reference selection in CD.

91


t, in the context of the start of the route s.

The inclusion of the reference to the district d0 is excluded by Rule 5. It

does not provide any information value to the hearer in the spatial context of s,

as s is covered by d0.

The next potential referent of finer granularity is d3. It does provide infor-

mation value in the given spatial context, and therefore represents a non-trivial

specification of the space containing the destination. It satisfies all the rules for

the selection of a reference for the destination descriptions.

While the reference to d32 is of finer granularity, and therefore provides more

precise specification of the destination t, it does not adhere to the principle of

relevance by increasing the cognitive effort of the hearer. The prominence of

any finer reference is lower than that of d3, and more topologically remote. A

reference to a finer reference does not provide the speaker with guarantees of

correct interpretation, as the finer the reference, the higher is the danger that it

will not be part of the spatial knowledge of the hearer. Thus, the reference to d3

should be selected to become reference r1 of the destination description, providing

relevant information to the hearer in the context of s and with consideration of

the structure of the environment.

In the hierarchical structure of districts in the set CD, the destination t of

the route routes,t is necessarily an ancestor of the reference r1 of the destination

description. This element specified by the initial reference covers the destination.

The destination is of finer granularity, deeper in the hierarchical structure of the

environment. Destination description must, therefore, include additional refer-

ences to specify the destination’s location in more detail. These references point

to districts that are subordinate elements of the reference r1. The selection of all

the child elements of the reference r does not, however, provide any guarantee

about the relevance of these references. All references included should satisfy the

principle of relevance, i.e. satisfy Rules 1-6. As hinted in Section 5.3, the selec-

tion of consecutive references is possible by a recursive evaluation of the selection

rules, with the consideration of a changing context.

Consider the specifications of context made in the model proposed (Sec-

tion 5.1). The condition of co-presence of the speaker and the hearer is central

for the specification of the context in which the selection of the reference occurs.

This co-presence need not be strictly physical, but may be merely projected.

Following the utterance of the reference r1, the spatial context considered by the

speaker changes. It is now defined by the reference r1. More precisely, the speaker

assumes that the district specified by r1 is accessed by the route. This region is

accessed at its periphery. The consecutive reference r2 will be interpreted from

92


this new spatial context.

The speaker assumes that this is the spatial context in which the hearer will

interpret the consecutive reference r2. Note that the speaker does not have means

to know exactly where the district r is accessed (as destination descriptions are

not prescriptive and the route is not communicated explicitly).

After reaching the first district of the route covered by r, the remaining part

of the route necessarily consists of districts fully covered by the district referred

to by r. Thus, the new spatial context s′ in which the consecutive reference of the

destination description (r2) is selected can be defined as a the first district along

the route routes,t covered by r. It is the first district of the remaining part of

the route. The notation subrouter will be used for the part of the route routes,t,

contained within the region of a reference r (subrouter = routes′,t). The new

context is not shared overtly between the speaker and the hearer, although they

both attempt to imagine a prototypical location, as a consequence of attempts

to be cooperative and relevant.

The identification of the consecutive reference of the destination description

then follows from the new spatial context s′. The process of selection of the

reference r2 in the context s′ is then identical to the process of selection of the

reference r1 in the context s. Rules 1- 6 thus apply for any spatial context, as

shown in Algorithm 3.

Algorithm 3: recDirs – recursive selection of district references for desti-nation descriptions

Data:routes,t: route, list of districts of finest granularity between start s anddestination t, defined as s and a trailing list of elements sx, where the lastelement is the destination t, (routes,t = s : sx);CD: Dataset–hierarchical partition of districts.Import: ref – function for the selection of the reference r.Result: r: list of references for the destination description of the route

from s to t.retrieve reference r1 (ref s t CD);1

case r1 = t2

return t;3

otherwise4

return r1;5

recDirs sx CD;6

To demonstrate the selection of consecutive references of the destination de-

scription, consider once again the example of the route122,322 (Figures 5.1 and 5.2).

The selection process of references for the route122,322 is illustrated in Figure 5.3.

93


Figure 5.3: Illustration of the process of selection of references for the destinationdescription of the destination d322 of the routed122,d322

. The gray triangles denotethe district considered in each step.

Once again, the selection of references consists of the evaluation of the or-

dered set of superordinate elements of the destination d322: [d0, d3, d32, d322].

After the retrieval of the first reference r1 = d3, the spatial context changes

from s = d122 to s′ = d341, as d341 is the first defining district of routed122,d322

within the spatial context of the reference r1 = d3. The consecutive reference

searched for must be relevant in the context of the sub-route of routed122,d322,

namely subroute3 = routed341,d322.

The next district considered for reference is d32. In the current spatial context

s′ = d341, the information value of the reference to d341 is low. d32 is neighbouring

the district d341. The reference to d32 does, at best, provide the hearer with

information one step closer to the destination. Any route within the spatial

context of r1 = d3, transiting through d34 must necessarily lead through d32. As

the reference to any other element of the granularity of d32 (i.e., d31 and d33)

is also missing, the consecutive reference must necessarily be within d32. As

the model of destination descriptions proposed provides a strict interpretation

of the principle of relevance, the hearer’s own spatial knowledge may be used to

substitute this information. As shown in following sections, the reference to other

types of elements of the city than districts may help reduce potential ambiguity

in such situations.

As the reference to d32 does not satisfy the rules specified, it is retained by the

speaker and a finer reference is sought in the set of the remaining superordinate

elements of the destination. As there is no further reference of intermediate

granularity available to the hearer in the given context, the reference to the

destination is included. The granularity of the spatial hierarchical partition CD

does not allow more detailed route directions of the route described.

The resulting sequence of references in the destination description for the

94

M. TOMKO 5.6. RULES FOR SELECTING LANDMARKS REFERENCES

route122,322 is:

d3 d322

The reference to only one superordinate element of the destination is appro-

priate in the context of the given hierarchical structure of the environment in

combination with the start and the destination of the route described.

As shown, the retrieval of consecutive references for destination descriptions

is governed by the application of the same rules as is the selection of the first

reference. Note that this is different to the pure selection of the branch containing

the destination in the spatial hierarchy. The principle of relevance acts as a

requirement to keep the amount of references to a minimum. Every new reference

included adds non-trivial information that can not be substituted from a-priori

spatial knowledge of the hearer to the resulting referring expression.

In the previous two sections, only a hierarchical partition of the environ-

ment into districts was considered. The inclusion of other types of hierarchical

structures, as well as the relevance-based inclusion of references to other types

of elements of the city from integrated experiential hierarchical datasets will be

considered in the following sections.

In the following sections, the basic principles for the selection of references for

destination descriptions in a hierarchical partition of districts are generalized to

accommodate for different types of elements in the city organized in hierarchical

structures of elements of different properties. These elements may be organized

in hierarchies of partitions or hierarchical rankings. The validity of the rules for

the inclusion of references in destination descriptions must be tested to adapt to

the characteristics of such hierarchies, as well as the properties of the elements

organized within them.

5.6 Rules for Selecting Landmarks References

As shown, neighbourhood is the fundamental topological relation driving the se-

lection of references for destination descriptions in a hierarchical partition of the

city into districts. Due to their duality with districts, the selection of landmark

references for destination descriptions is likely to be governed by the same prin-

ciples as the selection of district-based references.

The hierarchy of landmark reference regions, however, is a hierarchical order-

ing of spatial partitions. The landmark hierarchy is not a tree. Some landmarks

are present at consecutive granularities, and thus present their own superordinate

elements. Hence, the selection rules need an adaptation to be able to navigate

95


in a hierarchy with multiple inheritance and multiple appearances of the same

landmark.

Rules 1- 4 apply to both the hierarchy of districts (hierarchical partition) and

the hierarchy of landmarks (built as hierarchies of partitions of reference regions)

equally, as they verify that a destination description of a meaningful route is

requested. The application of Rules 5 and 6 to the hierarchy of landmarks has,

however, several shortcomings:

• The reference is always made to the landmark, and not to its reference

area. The identity of the landmark is unique even if its reference area

changes across granularities. Thus, it is possible that a reference is made

to a landmark that at is at the coarser granularity the ancestor of both

the current location and the destination. Rule 5 is therefore irrelevant for

landmarks.

• As a consequence of being organized in a hierarchy of partitions, it is im-

possible to test the neighbourhood relation between the reference regions of

landmarks of different granularities. There is no conceptual grounding of the

neighbourhood relation between point-like landmarks of different granular-

ities. Neighbourhood relations between the reference regions of landmarks

can therefore only be evaluated from the association information used dur-

ing the process of the formation of the hierarchy, contained in the Delaunay

triangulation.

The presence of a landmark across multiple granularities in the hierarchy may

lead to the selection of the same reference multiple times. The resulting destina-

tion description would then contain redundant references to the same landmark.

Furthermore, the rules proposed for the hierarchical partition of the environment

into districts are not ready to deal with multiple ancestors of an element. This

means that, at any granularity, the set of possible referents is not restricted to

one landmark. The implementation of Rules 5 and 6 must be modified to apply

across one step removed granularities. This results in a jointly applied Rule 7):

7. Element r of granularity g should not be neighbor with an element in Supers

of granularity g.

Furthermore, two new rules are proposed, enriched with mechanisms to deal

with redundant referents in the resulting destination description, as well as with

the selection among multiple possible referents:

96


8. Among possible referents, priority is given to the referents along the route.

If multiple landmarks satisfy this condition, the landmark closest to the

destination is selected.

9. If a landmark is referred to multiple times, remove all but one reference.

The underlying assumption is that the relevance of a landmark to the hearer

is higher if the landmark is en-route. Thus, in the context of landmark-based

destination descriptions, topological distance is not the only rule that applies

in the speaker’s choice of references. The consideration of the route context

is a plausible mechanism to select among multiple referents, especially at finer

granularities. The modified set of rules, including the Rules 7 and 8, is presented

in Algorithm 4 as the function ref ′.

The principles of selection of landmark references in destination descriptions

are formalized in the modification of the function recDirs from Algorithm 3 in a

process of filtering the resulting set of referents r. First of all, to accommodate

for the multiple possible ancestors in a hierarchy of partitions, the set Supert

becomes a list of lists of superordinate features of t. Each of these lists contains

the multiple possible ancestors of t at each granularity, instead of a single element.

The modified function recDirs′ is shown in Algorithm 5.

The application of the modified set of rules is demonstrated in the following

example of landmark-based destination description for the routed5,d1(Figure 5.4).

The specification of the route from the start s = d5 to the destination t = d1 is

based on the finest granularity (g) districts is as follows:

dg5, dg

4, dg3, dg

2, dg1

Due to the duality of landmarks and their reference regions—districts, a

landmark x of granularity k (lkx) is related to the district dx of granularity k (dkx).

A candidate set of landmarks, superordinate to the destination t = d1 can then be

identified. They are presented in order of decreasing (coarse to fine) granularity:

lg+46 ,

[

lg+37 , lg+3

2

]

, lg+22 , lg+1

2 , t

Note the multiple possible referents selected at the same granularity, as indi-

cated by the brackets. As the hierarchy of landmarks allow for multiple ancestors

of a landmark, multiple references are possible. The consideration of the con-

text provided by the route (routed5,d1) then applies. The references lg+3

2 , lg+22

and lg+12 to the landmark l2 is possible at multiple granularities. Furthermore, at

the granularity g + 3, the reference to the landmark l7 is an alternative to lg+32 .

97


Algorithm 4: ref’ – selection of the landmark reference r in the context ofs, t and routes,t.

Data: routes,t: route, list of landmark reference regions of finestgranularity between start s and destination t, defined as s and atrailing list of elements sx, where the last element is the destinationt, (routes,t = s : sx);

CL: Dataset–hierarchy of landmarks;s: current location (start reference region)t: destinationlgx: landmark lx of granularity g.Result: The reference r for destination description to t in the context of sConstruct candidate sets of superordinate reference regions of s and t,1

ordered by decreasing granularity (Supert = [ln, . . . t]) andSupers = [ln, . . . s]);Compare Supers, Supert;2

case (s, t /∈ CD)3


case (s = t)5


case (s, t = neighbors)7

Route too short for destination descriptions, switch to turn-based8

directionscase (Sups, Supt = neighbors)9

Route too short for destination descriptions, return t and switch to10

turn-based directionsforeach lgx : (lgx ∈ Supert) ∧ (∀lgy ∈ Supers, l

gx ∩ lgy = 0) do11

put lgx in list;12

order list in order of descending granularity;13

return list′ of landmarks lgx of coarsest granularity from list;14

foreach (lgx ∈ list′) do15

if (lx ∈ routes,t) then16

put lgx in list′′17

if list′′ 6= ∅ then18

order lgx in list′′ in order of appearance of reference regions of lx in19

routes,t;return r = last lgx in list′′;20

else21

return r = first lgx in list′;22

98


Figure 5.4: Illustration of the process of selection of landmark-based references forthe destination description for the routed5,d1

. The route is modelled as a sequenceof districts of granularity g. The references available for selection at granularitiesg + 4, g + 3 and g + 2 are displayed, including their respective reference regions.The resulting set of references selected for the destination description is shown(DD).

99


Algorithm 5: recDirs’ – recursive selection of landmark-based referencesfor destination descriptions

Data:routes,t: route, list of districts (reference regions) of finest granularitybetween start s and destination t, defined as s and a trailing list ofelements sx, where the last element is the destination t, (routes,t = s : sx);CD: Dataset–hierarchical partition of districts.Import: ref ′ – function for the selection of the landmark reference r.Result: r: list of landmark references for the destination description of

the route from s to t.retrieve list of references list (ref’ s t CD);1

case r1 = t2

return t;3

otherwise4

return rx;5

recDirs sx CD;6

keep a single entry of finest granularity for each rx ∈ list;7

While landmark l2 is en-route, i.e. its finest granularity reference region (d2)

is part of the route specification (d2 ∈ routed5,d1), landmark l7 is not en-route

(d7 /∈ routed5,d1). If no preference can be given through the consideration of the

route context, the selection of the reference is arbitrary. Landmark l2 is given

preference and is referred to in the destination description:

lg+46 , lg+3

2 , lg+22 , lg+1

2 , t

Landmark l2 is selected, through its reference regions, at consecutive gran-

ularities g + 3, g + 2 and g + 1. This process is described in Algorithm 4. The

application of the Rule 9 then follows, as described in Algorithm 5. It removes

duplicate references and returns the final, abbreviated sequence of references for

the destination description :

l6, l2, t

While the reference to a landmark is included by the speaker at the coarsest

granularity possible to minimize cognitive effort, it is interpreted by the hearer

at the finest granularity available in her or his a-priori spatial knowledge. The

destination is searched for in its vicinity, modeled here as reference region.

The resulting sequence of landmark-based references is shorter then the

equivalent sequence of references to districts in destination descriptions for the

same route. Before the application of the Rule 9, the sequence had five references,

100

M. TOMKO 5.7. RULES FOR SELECTING PATHS REFERENCES

as opposed to the final three references to landmarks (both numbers include the

destination). The different properties of landmarks allow for a brief set of refer-

ents in the destination descriptions, while preserving the relevance of the resulting

set of references. A reference to a landmark may be interpreted at multiple gran-

ularities. This property may be the reason why references to landmarks are so

frequently made by people, and why route directions and destination descriptions

with landmarks are considered useful.

The duality of landmarks with their reference regions allow for a flexible se-

lection and interpretation of an element as a landmark or district reference. Thus,

an explicit distinction is not necessary in the selection algorithm. Instead, the

semantic characteristics of the references may be considered. Thus, if a reference

region of a landmark is selected at a coarse level of granularity, and it covers an

area equivalent to an administrative district, the reference to the district’s name

is appropriate. District names represent a reliable common naming scheme in a

given city.

5.7 Rules for Selecting Paths References

Now, the hierarchical structure of paths is considered in an integrated manner

with districts and landmarks. Paths can be organized by prominence in a dis-

tinct hierarchy determined by the partial order of experiential prominence (Sec-

tion 4.4.3). Furthermore, they are integrated with and into the hierarchy of

districts through their structural properties, more specifically the relation of con-

tainment (Section 4.5).

The following notation is therefore introduced:

• A path pi is modeled as a sequence of connected finest granularity districts

pi = [dm, ..dn];

• The experiential rank value of a path pi is denoted as Ei;

• The set of all paths in the integrated hierarchy C is denoted as CP ;

• The mean value of the experiential rank of all paths in the hierarchy CP is

denoted as Ex.

• A path pi is said to be prominent if Epi≥ Ex. Such path can be noted

ppromi .

• The set of prominent paths together constitutes the experientially promi-

nent skeleton of the street network CpromP .

101


• The integration with the district hierarchy based on containment is based

on the structural relation of the path with the districts in the hierarchy

CD. The structural granularity of the path pi is defined as equal to the

granularity of the finest granularity district that fully covers the path pi. If

this granularity is g, the path pi can be noted as pgi .

The paths in CP represent the elements of the structure along which wayfind-

ers move. They can connect distant districts, and a reference to such paths can

therefore radically decrease the need for other references, especially if the path is

prominent. Only references to paths which connect the districts along the route

are referred to by the speaker. The set of such paths is noted Proute. Note the

difference from turn-based directions, where references to all paths constituting

the route need to be referred to, to unambiguously define the route to the hearer.

The application of the principle of relevance in the selection of path referents

requires the consideration of the experiential rank value Ei of the path, as well as

its structural granularity g. A reference to a path can be only made if the path

is prominent. The only exception is that of a direct connection of the current

spatial context and the destination of the route, when the speaker can refer to

the path directly (e.g.: “follow this path to destination.”).

Multiple paths can be available as possible referents along the route. The

preference order for the selection of referents, derived from the application of the

relevance principle, is summarized in the following rules:

10. Direct connectivity: the path providing direct connection between the cur-

rent spatial context and the destination should be selected, disregarding

whether it is prominent or not;

11. Prominence: the most prominent path from the set of alternatives should

be selected;

12. Structural granularity: structural prominence of paths allows to integrate

the references to paths with references to landmarks and districts in the

destination descriptions.

Note that the rules of the selection of path references for destination descrip-

tions also consider cognitive effect as the function of prominence (experiential

or structural) and cognitive effort as the function of topological distance (direct

connectivity). Due to the characteristics of paths, there is no need for measures

avoiding the selection of trivial references—a reference to a path directly connect-

ing the current spatial context with the destination is the most relevant reference

102

M. TOMKO 5.8. INTEGRATED DESTINATION DESCRIPTIONS

possible. Furthermore, this property provides an insight to the transition between

destination descriptions and turn-based directions (see Chapter 7).

The fundamental property of paths, namely the facilitation of connections

between two locations, requires the insertion of a district or landmark reference

after or before the insertion of the path referent. A reference to a path can never

stand alone, the wayfinder needs to receive information about either the direction,

or the extent to which to follow a path. The omission of such reference would

include inconsistency and ambiguity in the resulting directions. If the reference

to the district or landmark follows the reference to the landmark, it provides

both the information about extent and direction. If the reference to district or

landmark precedes the reference to the path, the direction is inferred (away from

the district or path). The extent has to be acquired from environmental clues by

the hearer. This usually occurs when the reference is made to a prominent path

directly leading to the destination.

5.8 Generation of Integrated Destination De-

scriptions

By the application of the principles above, path referents can be included into

district and landmark based destination descriptions in a recursive manner. The

set of paths Proutes,tis ordered in a sequence of their appearance along the route

routes,t. For every new spatial context s, s′, ... the paths from Proutes,tare evalu-

ated for their relevance as potential referents, as shown in Algorithm 6.

The simple combination of the topological distance, the hierarchical rank of

a spatial element and the context of the route, combined in set of rules, provide

means for a computational interpretation of the principle of relevance enabling

to select references for destination descriptions.

103


Algorithm 6: granularDirs – integration of district/landmark referenceswith path references pi in the context of the route routes,t.

Data: CP : Dataset–hierarchical ranking of paths; CD: Dataset–integratedhierarchy of districts/landmarks; routes,t: definition of the routefrom s to t

Import: recDirs′—reference selection function, returning the set ofreferences l1..n.

Result: The set of references pgi to paths, integrated with references ri to

landmarks/districts for the destination description of thedestination t in the context of s

Construct the set of path Proutes,t;1

Set r as the first reference provided by recDirs′;2

if |recDirs′| 6= ∅ then3

forall (pgi ∈ Proutes,t

), ordered by prominence in decreasing order; do4

Take first pgi ;5

case pgi ∩ t 6= 0 ∨ (pg

i ∩ s 6= ∅6

return pgi and t7

case granularity g of pgi ≤ granularity li8

Put ri = li;9

Set recDirs′ − li;10

Repeat granularDirs;11

case (pgi ∈ Cprom

P ) ∨ (g ≥granularity of li)12

Put pgi + li;13

Set recDirs′ − li;14

Set Proutes,t− pg

i ;15

Repeat granularDirs;16

otherwise17

recDirs′18

else19

Stop20

104

Chapter 6

Computational Implementation

of the Model of Destination

Descriptions

Haskell (Peterson and Chitil, 2005), a purely functional programming language

that enables implementation of an executable version of a formal model, with a

focus on the what instead on the how, is the programming language of choice

for the implementation of the model of destination descriptions presented in this

thesis. Programs implemented in Haskell profit of its properties: abstraction of

functions and types, genericity, polymorphism and overloading. Efficiency of the

code presented is neglected, although efficiency is gained by the lazy execution

paradigm of Haskell. Also, the focus has been on the demonstration of the el-

ementary properties of the model, which means that some properties necessary

for a real-life implementation were neglected. For instance, the implementation

presented requires pre-processing of the input data of Hannover (Chapter 4) into

a specific text-based format (Appendix B). Standard GIS data formats cannot

be used as inputs at this time. The focus on the properties of the model imple-

mentation is deliberate, as the adaptation to standard input formats can be done

through existing standard libraries.

6.1 Data Types

A Haskell program consists of functions, taking as inputs variables of various

datatypes. The model of the urban space in which the speaker and the hearer

interact and provide destination descriptions is modeled as set of objects of dif-

ferent types. The following main types are used: Path, District, and Object.

Note that District is the alternative structure to the definition of a reference

regions and districts, as well as landmarks and therefore no type Landmark is

105

CHAPTER 6. MODEL IMPLEMENTATION

needed.

Elements district and landmark are represented together in a unique struc-

ture of the abstract data type District. They are represented as a list of four

values: the identifier ObjectID, the granularity Level, name ObjectName and

the list of the subordinate elements Landmarks, which are the generators of the

reference regions (districts) of finer granularity, and the list of ObjectID of its

neighboring districts (Neighbors). Appendix B contains the data type definitions

along with the test dataset for central Hannover.

Elements of type path are modeled as abstract data types Path, represented

as a triplet (list of three values) of the path identifier (ObjectID), its constituent

districts (a list of ObjectID of the instances of districts), and the experiential

rank value of the path.

type Path = (ObjectID ,Districts ,Expvalue)

type District = (ObjectID ,Level ,ObjectName ,Landmarks ,Neighbors)

There is no explicit distinction between these types in the destination de-

scriptions, and thus both instances of types Path and District are treated as

instances of type Object in the integrated dataset. The constructors Area and

Street are used to construct such types. The distinct names are used for clarity

of the code.

data Object = Area ObjectID Level ObjectName Landmarks Neighbors

| Street ObjectID Districts Expvalue

Instances of datatypes Object are constructed from instances of datatypes

Path and District through the function consObject a, taking as an argument

an instance of any type and returning an Object (Appendix C, line 10).

class Elements a where

consObject :: a -> Object

createWorld :: [a] -> [Object]

The function createWorld a takes as an argument a list a of instances of

types District or path and returns a list of instances of type Object. From

there on, all functions are defined in a polymorphic, overloaded manner on het-

erogeneous instances of type Object.

6.2 Input and Output Specification

The inputs for the identification of the references for destination descriptions are

represented by the integrated hierarchical dataset World, and a list of objects

route, specifying the sequence of finest level districts defining the route between

the start and the destination. The input of the World is in-built in the program

106

M. TOMKO 6.3. MAIN FUNCTIONS

and is therefore not explicitly passed to the main function granularDirs of the

program as a parameter. Thus, the only parameter of the main function is route.

Note that the start of the route s and the destination t can always be derived

from the definition of route as its first and last element, respectively.

The dataset World is constructed by applying the function createWorld

to the list of paths paths and the list of landmarks and/or districts areas as

arguments (Appendix C, line 28).

world :: [Object]

world = createWorld(areas) ++ createWorld(paths)

The specification of the route provides the definition of the spatial con-

texts evaluated during the selection of the references by the speaker. The spatial

context is the only personalization parameter considered in this model. Finally,

one last parameter influences the selection of references: the specification of the

threshold experiential rank value, allowing the definition of prominent paths in

the given dataset. By changing this value, the size of the prominent paths dataset

can be increased or decreased. In the following, the mean value Ex is used, defined

as the variable eimean (Appendix B), following the discussion in Section 4.4.3.

The output of the main function of the implemented model is a list of instances of

type Object, representing the referents selected for the destination descriptions.

6.3 Main Functions

The model for selection of references for destination descriptions is implemented

as a series of filtering and ordering functions, passing the results of one function

to the next. The main function of the program is destDesc, taking as an input

a list a of Object and returning a list of the same type. The input list is the

specification of the route, the output is the ordered list of referents. The func-

tion is defined as follows (note the filtering nubBy (equalByName) which parses

the resulting set of references to remove any duplicate references)(Appendix C,

line 93):

destDesc :: [Object] -> [Object]

destDesc a = nubBy (equalByName)(destDescA (grdDist a) (reverse (routePaths a)))

destDesc a is an embedding function for the main processing function

destDescA (d:dx)(p:px) (Appendix C, line 97), taking as inputs the set of

district and landmark references resulting from the execution of the function

grdDist a and the function routePaths a, selecting all the paths occurring

along the route (note the reverse operator, to order the list of paths in the

descending order of prominence).

107


The function destDescA (d:dx)(p:px) implements the specification from

Algorithm 6.

destDescA :: [Object] -> [Object] -> [Object]

destDescA (d:dx)(p:px)

| length (d:dx) == 0 = error "no directions are possible for a route of

length =0"

| length [x|x<-( dirConectByProm (subroute d (route))), elem x (selectPaths

route)] /= 0 = d:[head [x|x<-( dirConectByProm (subroute d (route))), elem x

(selectPaths route)]]

| (fetchLevel (head (ordByLevelAsc (p:px))) <= (fetchLevel(head (ordByLevelAsc

(d:dx))))) = [d]++( destDescA (dx)(p:px))

| (fetchLevel (head (ordByLevelAsc (p:px))) == (fetchLevel(head (ordByLevelAsc

(d:dx))))) = (([d]++[p])++( destDescA (dx)(px)))

| otherwise = (d:dx)

6.3.1 Selection of District and Landmark References

First, the construction of the set of district and landmark based references is

explained. The constituent function grdDist of destDesc wraps around recDirs

and ensures the correct hierarchical ordering of its output. The function recDirs

(Appendix C, line 117) executes the recursive retrieval of district and landmark

references for the destination description (Algorithm 3). The filtering of references

to a single instance of each landmark (Algorithm 5) is consequently achieved by

the application of nubBy (equalByName) in the function destDesc, as described

earlier.

recDirs :: [Object] -> [Object]

recDirs (s:sx)

|getRef s t (s:sx) == t = [t]

|otherwise = (getRef s t (s:sx)):( recDirs (subroute (getRef s t (s:sx))

(s:sx)))

where t = last (s:sx)

A test at the beginning of the function recDirs ends the recursion if the

destination district is returned as reference. In this case, t is added to the result

set and the function ends. If the test is negative, a reference is retrieved and

the function recDirs runs again to find consecutive elements. This time, the

parameter route is represented by the result of the function subroute (Appen-

dix C, line 257), that determines the districts constituting the route in the area

specified by the previous reference. The districts are in the same sequence as in

the complete route.

subroute :: Object -> [Object] -> [Object] -> [Object]

subroute i r obj = [x | x <- r, testObject i (findSupersOrd x obj)]

In the function recDirs, the component function getRef s t (s:sx) ex-

ecutes the conditions for the identification of the reference, as specified in Al-

gorithm 4. Note that as the input dataset is based on the hierarchy of land-

108

M. TOMKO 6.3. MAIN FUNCTIONS

marks, Rules 5 and 6 do not apply (Section 5.6) and are replaced by the Rules 7

and 8. The main constituent function of getRef s t (s:sx) is the function

compHier sbranch tbranch route, which applies the rules on the sets of super-

odinate elements of s and t (Appendix C, line 137):

compHier :: [Object] -> [Object] -> [Object] -> Object

compHier sbranch [] route = error "no input of destination branch"

compHier [] tbranch route = error "no input of start branch"

compHier sbranch tbranch route = if length (t:tx) == 0

then error "change to turn based directions"

else

if length [x|x<-t, y<-s, isNeighbor x y || x==y ] /= 0

then compHier (concat(sx)) (concat(tx)) route

else

if length [x|x<-t, y<-route , fetchName x== fetchName y] >=1

then last [x|y<-t, x<-route , fetchName x== fetchName y]

else (last t)

where (t:tx) = groupBy (equalLevel)(ordByLevelDesc ([t | t<-tbranch ,

(testObject t sbranch)==False ]))

(s:sx) = groupBy (equalLevel)(ordByLevelDesc ([s | s<-sbranch , (testObject s

tbranch)== False ]))

This function also implements the consideration of the context of the route

in the selection of the appropriate reference among several possibilities. If a

reference to a landmark is possible, and the finest-granularity reference region of

this landmark is part of the definition of the route, the reference to the landmark

(or its reference region at the granularity considered) is given preference and

selected for the destination description.

6.3.2 Integration of Path References

The function destDescA integrates district and landmark references with refer-

ences to paths. When a reference to a path is included, not only the current

spatial context and the the previous district reference must be considered, but

also the hierarchical order of the possible paths referents.

Note the use of the function dirConectByProm in the function destDescA

(d:dx) (p:px), which is used in the last stages of the inclusion of path references

in the sequence of district and path references (Appendix C, line 170). It returns

a reference to the experientially most prominent path available when two districts

are directly connected. It takes as inputs an ordered set of paths occurring along

the route, provided as the result of the function routePaths.

This function orders paths along the route based on their relevance in a

given spatial context based on a combination of their order of appearance along

the route, their structural granularity (i.e., the structural prominence returned

by the function fetchMaxLevel) and experiential prominence. It compares paths

109


along the route based on the ordering function compByRouteCTX. Paths are or-

dered based on their structural relation with the route: the closer to the desti-

nation does the path appear along the route, the more relevant is the reference

to such path in the given context. The inputs of the function are two triplets

of values, describing the paths compared following the template: (path ID, ID

of first district of appearance along the route, ID of last district of appearance

along the route). The function elemIndex returns the position index of a district

in the sequence of the route (Appendix C, line 158).

compByRouteCTX :: (Object ,Object ,Object) -> (Object ,Object ,Object) -> Ordering

compByRouteCTX (a,b,c) (d,e,f)

|( elemIndex c route < elemIndex e route) = LT

|( elemIndex b route > elemIndex e route) && (elemIndex c route == elemIndex f

route) = LT

|( fetchMaxLevel a < fetchMaxLevel d) = LT

|( fetchExp a < fetchExp d) = LT

|(( notElem (fetchName b) (grdDistShow route)) || (notElem (fetchName c)

(grdDistShow route))) && ((elem (fetchName e) (grdDistShow route)) || (elem

(fetchName f) (grdDistShow route)))= LT

| otherwise = GT

6.4 Model Verification

To verify the model, routes of various lengths and complexities across a test area

of central Hannover were constructed. Consecutively, destination descriptions

for these routes were generated, and their adherence to the rules specified was

verified. The content of the destination descriptions was assessed by comparing

the resulting sets of references with the characteristics of destination descriptions

summarized in Section 3.6. Note that the example of Stephanie arriving at the

airport and traveling to the center of Hannover could not be tested due to the

limited dataset available. The following principal characteristics of destination

descriptions were sought:

• Consistency: the resulting combination of references must create an unam-

biguous specification of the destination, thus resulting in a referring expres-

sion.

• Well-formedness: the destination descriptions should not have redundant

references, and each consecutive reference should provide relevant informa-

tion in the context of the previous one.

• Brevity: the resulting destination descriptions should combine integrated

references to heterogeneous types of elements in order to achieve relevance

and brevity. Thus, the reduction of length in comparison to homogeneous

110

M. TOMKO 6.4. MODEL VERIFICATION

destination descriptions and turn-based directions generated by a Web ser-

vice was sought.

• Content: The selection of relevant references should be dependent on the

hierarchical structure of the environment in the proximity of the destination

and not on the route imagined by the speaker. The assessment of the

plausibility of the content of the destination descriptions is based on the

individual judgment and the consultation with a local expert. Plausibility

is desired, but remains subjective.

Note that the combinations of references retrieved were tested against general

characteristics of destination descriptions generated for a hearer with inferred

extent of a-priori spatial knowledge of the environment. The spatial knowledge

of the speaker was modeled in an experiential hierarchical dataset. The sets

of references retrieved are therefore influenced by the content and quality of the

dataset (limited extent, assessment of the properties of landmarks). For instance,

the complete path street network available was considered for the references, as it

would be if the hearer was a pedestrian. Central Hannover has, however, a large

pedestrian zone. As mentioned earlier, the shared functional perspective of the

speaker and the hearer on the structure of the environment is a requisite for an

appropriate selection of references for destination descriptions and the means of

transport used must be considered. It is therefore likely that the set of references

provided by a local to a taxi-driver would consider the accessibility of the different

parts of the city by considering the mode of transport.

This section is structured as follows: first, sets of district and landmark

based destination descriptions are retrieved and their content verified against the

specification of the rules for selection of district and landmark-based destination

descriptions (Section 6.4.1). Consecutively, the path references are considered

in the integrated model, and the results assessed for plausibility (Section 6.4.2).

The behaviour of the model is discussed on several model test cases. Additional

test cases can be found in Appendix D.

6.4.1 Test of District and Landmark-Based Destination

Descriptions

In the following example, the process of identification of district and landmark-

based referents for a route from the Universitat Hannover to the Staatstheater-

Oper (Figure 6.1) is demonstrated. A direct route from the Universitat Han-

nover to the Staatstheater-Oper has been defined by a sequence of districts (the

111


sequence contains districts of Level 2, hence the prefix h2 of the respective

ObjectID):

route = [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H03PO3Z , h2_H04SBR1 , h2_H01BHXG ,

h2_H01P2HN , h2_H04PTS0 , h2_H01FM8E]

Rathaus

KatasteramtStaatstheater-Oper

Universität Hannover

Figure 6.1: The route (highlighted in gray) between the Universitat Hannover(H097TLK) and the Staatstheater-Oper (H01FM8E), composed of level 2 refer-ence regions. Referents identified for destination descriptions are labelled.

The route generated cuts across the city and intersects the reference regions of

several global landmarks. These are landmarks that are not directly en-route,

their reference regions cover parts of the route. Applying the rules for selection of

district and landmark-based references in destination descriptions, the following

references are retrieved for the first definition of the route:

directions = [H06Y0NB , H04PTS0]

In common names, the references to the following landmarks were made:

directions = [Rathaus , Katasteramt]

These results can be interpreted as the following destination descriptions

generated by a local: A:“Where is the Staatstheater-Oper?”

B:“. . . in the direction of the Rathaus, next to the Katasteramt.”

112


The destination is found in the proximity of the first landmark specified, the

global landmark Rathaus (H06Y0NB). The context of the route is restricted to the

general area specified by the reference region of the landmark, and consecutive

references of finer granularity are provided (the Katasteramt (H04PTS0)). In

this manner, the destination descriptions proceed from a general reference to a

landmark with a reference regions covering major parts of the city to a more local

landmark. Global landmarks may have no spatial overlap with the route or the

destination as such, but they provide a description of the destination from the

approaching direction of the wayfinder. The route description could be completed

with the reference to the destination itself, Staatstheater-Oper (H01FM8E).

Due to the duality of landmarks with districts, the following expression to

districts can be considered equivalent (see Appendix A):

directions = [Hannover Mitte , Kroepcke]

which translates in the following destination descriptions: B:“. . . in the Mitte,

close to the Kropcke.” Note that Mitte is literally Centre in German. The

Rathaus is used in its function of a global landmark, and as the reference regions

of the Rathaus over coarser granularities cover Hannover Mitte, the reference to

the district or to the landmark can be used interchangeably in the verbalization

of the destination descriptions. Also note the verbalization of the references in

natural language need not follow the ordering retrieved by the model. Indeed, in

this example, the ordering is reverse. The resulting set of references represent a

plausible content of a destination description for a route from the University to

the Opera.

The route was generated for a pedestrian trying to reach the Staatstheater-

Oper as directly as possible. A different mode of transport preferred by the hearer

can change the inference of relevance of the references by the speaker. While a

different functional perspective on the path network is not considered, the route

considered by the speaker may influence the selection of references, due to the

preference of en-route landmarks.

For example, the suggested route from the Universitat Hannover to the

Staatstheater-Oper provided by Google Maps directions service is complex, as

the Staatstheater-Oper is in the pedestrian zone that has to be avoided (Fig-

ure 6.2 and 6.3). The resulting turn-based directions contain eleven references.

The wayfinder will approach the Staatstheater-Oper from a different direction.

The route generated by Google Maps has been translated into a district-based

definition of the route, and used as input to generate destination descriptions.

route(google) = [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H05V43Q , h2_H03PO3Z ,

h2_H03PO1O , h2_H03Q6S0 , h2_H01O23Z , h2_H04PTS0 , h2_H01FM8E]

113


Figure 6.2: Turn-based route directions for the route between the UniversitatHannover and the Staatstheater-Oper as provided by ©2007, Google Maps. Notethat the directions generated are for the Movenpick Restaurant in front of theStaatstheater-Oper main entrance, a point-of-interest contained in the GoogleMaps database.

Figure 6.3: Map of the route between the Universitat Hannover and theStaatstheater-Oper (©2007, Google Maps).

114


The references retrieved for the route suggested by Google Maps are identical to

those retrieved for the direct, pedestrian route from the Universitat Hannover to

the Staatstheater-Oper:

directions(google) = [H06Y0NB , H04PTS0]

This shows how the content of purely landmark and district-based destination

descriptions is primarily influenced by the granular structure of the city in the

proximity of the destination, instead of the route considered by the speaker. Of

course, the preference given to local en-route landmarks influences the content

of the resulting destination descriptions, but the primary influence is that of the

overall structure of space. The resulting set of references was assessed as plausible

by a local expert.

Consider a different route, in the reverse direction from the centre of the city

to the outskirts of the area covered by the dataset. A destination description of a

route from the Staatstheater-Oper to the Universitat Hannover consist of a single

reference, identical with the destination: Universitat Hannover (Appendix D,

example 4). It is not possible to find a more prominent landmark than Universitat

Hannover itself in this part of the city. If the wayfinder is not satisfied with the

direct reference to the University main building, only turn-based direction will

satisfy their information needs. Such a wayfinder has no sufficient a-priori spatial

knowledge of Hannover to be able to process inferential destination descriptions.

Again, a route from Staatstheater-Oper to the Universitat Hannover, gener-

ated by Google Maps was used as input for the model of destination descriptions.

This route avoids the centre of the city once again, and is therefore complex and

its turn-based directions consist of eight references. The resulting set of refer-

ences provided by the model of destination descriptions implements is, however,

unchanged: Universitat Hannover (Appendix D, Example 5). This shows how

the size of destination descriptions is dependent exclusively on the hierarchical

structure of the environment in the proximity of the destination. The content

may change if a different route is selected, due to the consideration of en-route

landmarks, but the overall destination description remains constant.

Consider an extension of the route to the Institute for Chemistry at the

University (H05MF0G). It is one of the buildings adjacent to the very promi-

nent main building of Universitat Hannover (H097TLK). The route taken covers

the same suburbs as the route from the Universitat Hannover building, with an

extension to the Institute (Appendix D, Example 6):

route = [h2_H01FM8E , h2_H04PTS0 , h2_H01P2HN , h2_H01BHXG , h2_H04SBR1 , h2_H03PO3Z ,

h2_H074YH2 , h2_H063YJC ,h2_H097TLK , h2_H05MF0G]

115


The resulting references retrieved by the model of destination descriptions

presented consist of only two referents: the Universitat Hannover (H097TLK)

and the building of the Institute for Chemistry (H05MF0G):

directions = [H097TLK , H05MF0G]

This destination description contains the reference to the prominent land-

mark, Universitat Hannover, followed by the reference to the Institute itself,

which is considered optional. Note that the reference region of level 2 of Uni-

versitat Hannover (h2 H097TLK) is part of the route definition and Universitat

Hannover is thus a local landmark.

While the resulting sets of district and landmark-based references was as-

sessed by a local expert as plausible, specific routes such as the one from the

Staatstheater-Oper to the Institute for Chemistry do not provide guarantees

of low cognitive effort. Such routes are influenced by the proximity of a low-

prominence destination close to a prominent landmark. If environmental clues

are not present, it may be difficult for the hearer to identify the destination. This

drawback will be addressed by the integration of paths in the model, as shown in

the following section.

The resulting sets of references found for all combinations of routes are brief,

and provide relevant information based on the assessment of the urban structure

and the context of the route. Due to the hierarchical structure of Hannover, in

general all routes from the centre to the periphery (i.e., from reference regions

hierarchically subordinate to the Rathaus to reference regions subordinate to the

Universitat Hannover) have shorter destination descriptions than routes in the op-

posite direction. The hierarchical structure of the landmarks within the reference

regions of the Rathaus is more complex, and thus the destination descriptions

require more referents for a unique identification of the destination.

6.4.2 Test of Destination Descriptions with Paths

The integration of references to paths may further improve the consistency of

the content of the references retrieved by matching references to local landmarks

with references to paths. The integration of references to paths is now studied on

the example routes already presented, as well as on some additional examples.

Consider again the example of the routes from Universitat Hannover to the

Staatstheater-Oper. Based on the route specifications, new sets of references are

retrieved, this time based on the integrated hierarchies of districts, landmarks

and paths. The first example applies for the specification of the direct route from

the Universitat Hannover to the Staatstheater-Oper (Figure 6.4) (Appendix D,

116


Example 7).

route = [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H03PO3Z , h2_H04SBR1 , h2_H01BHXG ,

h2_H01P2HN , h2_H04PTS0 , h2_H01FM8E]

The resulting set of references identified by the implemented model of destination

descriptions is as follows:

directions = [H06Y0NB , H04PTS0 , N01FUH0]

or, in common names:

directions = [Rathaus , Katasteramt , Standehausstrasse]

which translates in the following destination descriptions (The reference to Rathaus

is replaced by the reference to its dual reference region, Hannover Mitte): B:“. . . in

Hannover Mitte, take the Standehausstraße from the Katasteramt.”

Note that the street Standehausstraße (N01FUH0) is not a prominent street

by experiential rank value. It is, however, the most prominent street directly

connecting the Katasteramt with the destination of the route, the Staatstheater-

Oper (Figure 6.4). As it is the last reference of the destination description, the

speaker relies on environmental clues that this street will be identified. Further-

more, a prominent local landmark, the Katasteramt, is at one of its extremities.

The resulting destination description leads the wayfinder closer to the destination

than the pure district and landmark based set of references. The integration of

multiple types of elements of the city allows generating destination descriptions

of finer spatial resolution. Thus, the integration of multiple types of references

allows retrieving a set of references providing less ambiguity to the hearer, by

better adapting to the structure of the environment. By integrating references to

paths, the speaker may provide destination descriptions that are not patronizing,

but at the same time provide the hearer with a general guidance on the approach

to the destination.

The inclusion of references to paths considers en-route paths. Therefore,

the content of the resulting destination descriptions may be prone to change

depending on the route considered by tyhe speaker. The set of references retrieved

for the route generated by Google Maps and shown in Figure 6.3 is once again

used for comparison. In this case, the resulting set of retrieved references is

identical (Appendix D, Example 8):

directions = [H06Y0NB , H04PTS0 , N01FUH0]


directions = [Rathaus , Katasteramt , Standehausstrasse]

The identical content of the two destination descriptions generated by the

consideration of two different route specifications shows that the structure of

117


Staendehausstrasse

Rathaus

Staatstheater-Oper

Universität Hannover

Figure 6.4: References selected for the destination description for the route fromUniversitat Hannover to the Staatstheater-Oper. The route is shown in gray shad-ing. The references retrieved: Rathaus, Katasteramt and the Standehausstraßeare labeled.

118


the environment in the proximity of the destination largely determines even the

content of destination descriptions containing references to paths. In general,

only the structure of the final parts of the route impacts on the selection of path

references. This will be demonstrated on the following examples.

Consider once more the direct route from the Staatstheater-Oper to the

Institute for Chemistry. The references retrieved are as follows:

directions = [H097TLK , N01FUPO]

which can be translated into common names as follows:

directions = [Universitaet Hannover , Im Moore]

which translates in the following destination descriptions : B:“. . . from Universitat

Hannover, follow Im Moore.”

Note the different content of the destination descriptions considering refer-

ences to paths, districts and landmarks, as opposed to the set previously gen-

erated without the consideration of paths (Appendix D, example 6). While the

previous example contained only the reference to the Universitat Hannover, with

the optional reference to the destination itself, the newly generated destination

descriptions omit the reference to the destination, but include a reference to the

street Im Moore. This street directly connects the prominent en-route landmark

Universitat Hannover with the destination.

The destination description was judged by the local expert as satisfying the

information needs of a wayfinder with a-priori spatial knowledge of Hannover.

Note that once again, the preference is given to a directly connecting street,

instead of a prominent one (and indeed, there is no prominent street connecting

Universitat Hannover with the Institute for Chemistry). The integrated set of

references provides a destination description requiring less cognitive effort from

the hearer, and is thus more relevant than the purely district and landmark

based destination description. This is easily verifiable, as the two sets are of

equal length, but the integrated set is more information rich.

Finally, consider one last example, a route leading from the Allianz-Hochhaus

to the Katasteramt (Figure 6.5):

route = [h2_H05V43Q , h2_H03WTT1 , h2_H01F6M0 , h2_H04PTS0]

The references retrieved for the destination description are:

directions = [H04PTS0 , N01FUJS]


directions = [Katasteramt , Karmarschstrasse]

119


Karm

ars

chstr

asse

Katasteramt

Allianz-Hochhaus

Figure 6.5: References selected for the destination description for the route fromAllianz-Hochhaus to the Katasteramt. The route is shown in gray shading. Thereferences: Katasteramt and the Karmarschstraße are labeled.

120

M. TOMKO 6.5. OBSERVATIONS OF THE MODEL OUTPUTS

which translates in the following destination descriptions : B:“. . . take Karmarschstraße

to the Katasteramt.”

Note that a local of Hannover is more likely to substitute the reference to the

landmark Katasteramt by its reference region, the Kropcke, with the resulting

destination description: B:“. . . take Karmarschstraße to the Kropcke.”

The reference to the Karmarschstraße is appropriate, as it is a prominent

street, directly connecting the districts along the route specified to the Kataster-

amt, the seed landmark of the reference region known as Kropcke. This is a

combination of references likely to be used by any local in Hannover for the route

specified.

The examples discussed show how the inclusion of heterogeneous references

allows more relevant destination descriptions to be constructed. The integration

of references to paths may also result in destination descriptions of smaller size

of the resulting set of referents, compared to the set of districts and landmark-

based references. The model presented demonstrates that it is possible to select

references for the construction of plausible and relevant destination descriptions

based purely on the analysis of the integrated hierarchical urban structure and

the recursive evaluation of the spatial context of the hearer.

6.5 Observations of the Model Outputs

The references selected by the implemented model provide a plausible content of

destination descriptions for people with a-priori spatial knowledge. A-priori spa-

tial knowledge is necessary for the understanding of the meaning of the references

provided. The inclusion of different types of referents allows the model to adapt

to the spatial structure in the vicinity of the destination, and allows it to select

references relevant to the hearer.

The results of the model show the following characteristics:

• The number of references is not dependent on the complexity of the route

retrieved, but on the complexity of the hierarchical organization of the

environment in the proximity of the destination;

• The number of references retrieved is small and shows similar patterns (dis-

cussed below);

• The alteration of the threshold defining the set of prominent paths in the

destination descriptions does not influence the overall length of the results

significantly, but changes the contents pattern of the results.

121


In the examples studied, the length of generated sets of references in the

integrated hierarchical spatial data structure of Hannover ranged between 2 and

3, or 4 if the reference to the destination itself is included. Note that this inclusion

is not necessary, as it is used in the hearer’s request as such. Its inclusion may

therefore be a pattern of language features or cultural habits. The length of the

routes varied in length from two to more than nine level 2 districts. The depth

of the granular structure of space consisted of six hierarchical levels used in this

experiment. Thus, the resulting sets of references represent a small proportion

of the superordinate elements of the destination in every test case. Note that

a deeper hierarchical structure need not necessarily lead to longer destination

descriptions . The pattern of the parent-child relationships in the structure must

be analysed to infer the average length of the resulting destination descriptions.

The patterns identified in the integrated sets of references for destination

descriptions tested are shown in Table 6.1.

Table 6.1: Patterns in sets of references in destination descriptions (D–district,LG–global landmark, LL–local landmark, P–path, P Prom–prominent path).

Pattern 1 2 3 4 51st reference D/LG D/LG LG = LL LG = LL P Prom

↓ ↓ ↓ ↓ ↓2nd reference D/LL P Prom P Prom P t

↓ ↓ ↓ ↓3rd reference P D/LL

Optional reference t t t t

In Pattern 1, the reference to a global landmark or significant part of the

city narrows down the search space for the next reference. It assists the hearer

to interpret the following reference correctly. In the reference region of the global

landmark, the next reference is located. This reference is made to a local land-

mark (or its reference region). This landmark is thus en-route. The selection of a

reference to a local path between the local landmark and the destination signifies

that no prominent path connects the destination with the reference region of the

local landmark selected previously.

If a prominent path is found in the area specified by the global landmark

(Pattern 2), it is used to guide the hearer to the local landmark, in which vicinity

the destination is found. The reference to the destination found in the vicinity of

the local landmark is optional.

Patterns 3 and 4 occur when the destination is in the direct vicinity of the

global landmark. Then, the global landmark is en-route and serves as a local

122

M. TOMKO 6.5. OBSERVATIONS OF THE MODEL OUTPUTS

landmark as well. A reference to a path (prominent or not) is then inserted to

guide the wayfinder toward the destination.

Pattern 5 occurs in cases when the destination is located in proximity (en-

route) of a prominent path. The reference to the destination is than necessary, to

provide the wayfinder with a stop criterion. It is assumed that the hearer will be

able to identify the destination visually. Note that this model does not provide

means to insert other types of metric information than a reference to a spatial

feature.

The only means to alternate the selection of the references in the model is the

consideration of a different threshold of experiential rank value, thus enlarging

or reducing the set of prominent paths in the city. This does not dramatically

change the length of the resulting sets of references, but merely changes the

balance between the district/landmark references and references to paths. If the

threshold is lower, the balance shifts from district and landmark references to

references to paths, and leads to brief destination descriptions. Patterns 3, 4 and

5 become more frequent. If the number of prominent paths in the city is low, the

resulting sets of referents exhibit Patterns 1 and 2.

123


124

Chapter 7

Conclusions

7.1 Summary

The communication of spatial information has been the object of intensive study

of cognitive scientists, linguists, spatial information scientists and geographers,

computer scientists and researchers in the field of human-computer interaction.

Most of this research was based on the classical model of communication pre-

sented by Shannon and Weaver (1949). A common example is turn-based route

directions, encoding every action the recipient of the information needs to unam-

biguously follow the route and reach the destination. Classical communication

theory fails, however, to explain the communication of spatial and non-spatial in-

formation among people. While messages exchanged in everyday communication

and dialogs contain only a small part of the information necessary to perform a

task required, people receiving this information are able to interpret the meaning

conveyed.

Pragmatic information theories have been devised by linguists to explain

this observed discrepancy, but remained largely neglected by researchers in the

field of spatial information communication. Among notable exceptions are the

works of Frank (2003) and Worboys (2003). These works point to the importance

of a-priori information as an important part of the context in which the hearer

interprets the message received. Furthermore, Dale et al. (2005) introduced the

concept of referring expression generation to the generation of route directions.

These works were based on the interpretation of Gricean conversational maxims

(Grice, 1989). Among pragmatic information theories, the relevance theory of

communication has recently gained prominence by its ability to explain in a

plausible manner the inferential interpretation of the meaning of messages in

relation to the context of their communication.

Spatial information is often an essential part of information communicated in

125

CHAPTER 7. CONCLUSIONS

situations where correct interpretation of the meaning of the message is required

to assure the safety or integrity of people. Furthermore, irrelevant information

negatively influences the processing time required by the hearer to understand

the meaning of the information provided. The importance of communicating

only relevant information is thus obvious. Until now, however, application of the

relevance theory to the problem of communication of spatial information has not

been attempted.

This thesis applies the principles of relevance communication theory to the

problem of generation of destination descriptions. It introduces an inferential

model of destination descriptions for people with a-priori spatial knowledge of the

environment. This wide group of users consists of people frequently experiencing

a limited urban environment, e.g. a city. The formal model is demonstrated in a

computationally executable implementation, selecting references for destination

descriptions from a hierarchical dataset. Based on the structure of the dataset

and the definition of the route from start to the destination, the model selects

references for inclusion in the destination descriptions. The references provide

overview information about the destination of the route. Hearers interpret the

meaning of the references provided by relating the information received to their

a-priori spatial knowledge. The understanding of the meaning of the references

is performed by relating to the spatial context in which the references are pro-

vided. The outputs of the model satisfy the characteristics of human-generated

destination descriptions.

Together with the relevance-based model of destination descriptions, a cogni-

tively motivated approach to the construction of integrated hierarchical datasets

of landmarks, districts and paths is presented, allowing the speaker to assess the

relevance of a reference to individual spatial elements of the environment for the

hearer. Such hierarchical datasets integrate in a tight structure the experiential

hierarchies of heterogeneous spatial elements. The structure of the environment

is a significant determinant of the content of destination descriptions, and as such

may compensate for the lack of mutual background knowledge of the profiles of

the communicators.

7.2 Main Contributions

This thesis presents an analysis of the principles by which the structure of the

environment determines the content of destination descriptions exchanged in in-

ferential communication. It shows that the hierarchical structure of the environ-

ment, and the spatial context in which the destination descriptions are commu-

126

M. TOMKO 7.2. MAIN CONTRIBUTIONS

nicated, are strong determinants of the content of such destination descriptions.

Individual spatial knowledge is not considered in the model proposed. The main

contributions of the research presented in this thesis can be classified in the fol-

lowing broad categories:

• An operationalization of the relevance theory, or more precisely of the prin-

ciple of relevance in the spatial domain. This interpretation is based on

the interpretation of the concepts of cognitive effort and cognitive effect in

the spatial communication domain. The relevance of a spatial element in a

given context is defined in terms of its relative prominence, distance, and

the spatial context of communication.

• A cognitively motivated model of inferential destination descriptions, pre-

senting a formal approach to the selection of references for destination de-

scriptions for hearers with a-priori spatial knowledge. Destination descrip-

tions are presented as a special case of spatial referring expressions. The

correct interpretation of the references requires the consideration of the

spatial context in which they are uttered.

• The introduction of the concept of integrated, experiential hierarchies of

urban environment. Hierarchical datasets organized as experiential hierar-

chies provide a cognitively inspired means of ranking spatial elements by

the inferred perception of their prominence.

The operationalization of the principle of relevance is based on the analysis of

the principles of human spatial cognition. The principles governing the acquisition

of human spatial knowledge and the subsequent organization of the elements

of this knowledge were analysed. These principles were consequently applied

to infer the cognitive effort and cognitive effect of retrieving elements of this

spatial knowledge during communication. Furthermore, the relative distance from

the potential referent was linked to the cognitive effort required to process the

reference.

Based on the operationalization of the principle of relevance, a formal model

of selection of references for destination descriptions was proposed. The model

satisfies requirements for cognitively ergonomic communication of spatial knowl-

edge:

1. The resulting set of references is short, and does not increase linearly with

the length and complexity of the route. The inclusion of several types

of referents further decreases the cognitive workload they impose on the

127


wayfinder. Thus, destination descriptions satisfy the requirements of low

cognitive load.

2. Destination descriptions satisfy the requirements of good route directions

(Allen, 2000) by being orderly and by increasing the amount of referents in

the proximity of the destination (Allen, 1997). The content of destination

descriptions is restricted to the minimal relevant set of references uniquely

describing the destination. Destination descriptions are thus not patroniz-

ing, as they do not force the wayfinder to follow a prescribed route. This

is in contrast to turn-based route directions, which require the wayfinder

to follow the route planned by the provider of the directions. Destination

descriptions are thus suited for wayfinders with a-priori spatial knowledge.

As a consequence, Allen’s requirement for the inclusion of delimiters, such

as landmarks on decision points is not satisfied. This knowledge is supplied

by the wayfinder as such.

3. The model generates longer descriptions (consisting of more referents) for

less prominent destinations. It has been shown that landmarks judged less

well known by locals have been shown to be described by more detailed refer-

ring expressions (Lau and Chiu, 2001). Furthermore, individual familiarity

with spatial objects is highly correlated with the judgment of familiarity

of others (Fussell and Krauss, 1991; Lau and Chiu, 2001). The power-law

distribution of the experiential rank values in experiential hierarchies hints

at why the estimates of familiarity of others is so highly correlated, and

allows for a qualified estimate of general familiarity with an element of the

city.

4. The application of the simple rules implemented adheres to the heuristics

of relevance as stated in the relevance theory of communication. The inter-

pretation following the path of least effort is preferred, as is the case in our

model. A plausible behavior of the model is the result.

The model proposed focuses on the information needs of locals—people with

a-priori spatial knowledge of an environment. This does not mean, however, that

the generated destination descriptions match exactly the spatial knowledge of an

individual, or that a specific local would provide the same destination descriptions

for any route in the same context. The references selected using the model of

destination descriptions resemble to the set of references likely to be provided by

locals, and should be unambiguously interpreted by locals.

An analysis of the emergence of experiential hierarchies in the mental rep-

resentations of humans results in a model of integrated experiential hierarchies

128

M. TOMKO 7.3. DISCUSSION

applied to construct hierarchical datasets of urban environment. It is argued that

structural properties of the environment are paramount in the hierarchization of

the individual types of the elements of the city, and that these hierarchies are

interrelated.

The main properties of integrated experiential hierarchies are:

1. Construction of a cognitively motivated partial-order rankings for different

elements of the city with element-dependent properties of the partial-order

establishing the hierarchy. The result is a set of individual hierarchies with

tree-like, lattice and frequency ordering structures. The ranking of the

elements in a hierarchy is based on the estimate of prominence.

2. A novel, cognitively motivated measure to compare the relative prominence

of streets in street networks, called Experiential rank.

3. A method to integrate heterogeneous hierarchies in a unified structure, al-

lowing an interchangeable selection of different types of elements of similar

prominence.

An integrated experiential hierarchy of the environment is constructed for a

sample study area. The resulting dataset is used for the selection of references

for destination descriptions and tested for the city of Hannover.

7.3 Discussion

7.3.1 Cognitive Workload and Destination Descriptions

Destination descriptions are a specific form of spatial communication, combining

the properties of route directions and place descriptions. Destination descriptions

are provided to hearers that have a-priori spatial knowledge of the environment.

Destination descriptions are meant to be retained in the memory of such hearers.

Despite differences between various cognitive tasks, the span of short-term

memory was reported to equal approximately seven information items, with a

spread of two (Miller, 1956). Consecutive studies indicate even slightly lower

numbers (Cowan, 2001). Miller furthermore suggests that this capacity can be

increased if the items to be remembered are of different types. Thus, diversity

helps to increase the short-term memory span, with a lower increase with every

added type of item.

It appears that the length of destination descriptions is related to the depth of

the experiential hierarchies, not to the length of the routes. From administrative

129


classifications of districts to road classifications, the number of granularities in

which the respective domains of man-made phenomena are classified is relatively

small. For example, the hierarchy of elements of type district has a greatest ele-

ment, the city itself, and a theoretical least element—a parcel at a street address

level. Between these extremes, the hierarchy is organized in a small number of

levels, be it landmark reference regions (in the case presented with a depth of six

levels) or administrative partitions such as building block, quarter, and suburb.

A large number of elements can be contained in a granular system that can be

represented as a balanced tree of seven granular levels. A binary classification

of seven levels can contain a hierarchically ordered set with 64 elements at its

finest level. This is a large number of elements contained in a narrowest possible

hierarchy.

Thus, a theoretical destination description with references of each granularity

would contain seven references. While highly redundant, this extent still fits

in the short-term memory span of the hearer. While the number of references

in turn-based directions grows linearly with the complexity of the route, the

number of references in destination descriptions grows only logarithmically, being

proportional to the depth of the hierarchical structure of space considered.

In the model proposed the references selected do not represent the selection

of a complete branch of superordinate elements of the destination. Instead, the

recursive selection of references re-evaluates the changing spatial context in which

every consecutive reference is selected. The relation within integrated hierarchies

of the different elements of the city allows for even more efficient transition be-

tween the individual hierarchies of the different types of references, leading to an

efficient selection of references based on the structural properties of the spatial

features of different types.

When analysing the structure of destination descriptions provided by humans

in everyday conversations, several common characteristics are revealed (Tomko

and Winter, 2006a). The apparent reduction in information conveyed is mani-

fested by the lesser length of the destination descriptions, measured by the num-

ber of references to spatial elements. The length of the provided destination

descriptions correlates with the number of types of elements of the city referred

to.

The reduction of the number of references in destination descriptions is not

the only reason for their lower cognitive complexity when compared to turn-

based directions. The inclusion of multiple types of references extends the span

of short-term memory Miller (1956). The occurrence of three types of references

in the route directions in average points to the possibility to encode a complex

130

M. TOMKO 7.3. DISCUSSION

and lengthy route in an utterance that fits the short-term memory span of the

hearer. The diversity of references enables to increase this span sufficiently to

accommodate the increased information content. The three types of elements of

the city tested offer a diversity of referents enabling to construct efficient route

directions of considerable length, even in complex urban environments. Rele-

vance based selection grounded in hierarchical granulation of spatial features into

chunks, together with the alteration of the type of references, provides efficient

means to increase the amount of information to be communicated.

7.3.2 Reliability of Inference of Common Spatial Knowl-

edge

As shown, the shared context between the hearer and the speaker in inferential

communication is a major influence on the choice of referents. But how does

the selection of referents from experiential hierarchies influence the success of

communication? In Chapter 4, the distribution of the experiential rank values in

the urban network, following a power law, was shown. Similarly, the total number

of landmarks and districts is exponentially higher than the number of prominent

landmarks and districts. This is understandable, as prominence is a function of

salience and uniqueness, i.e. the function of rarity of a phenomenon.

If the hierarchies of the spatial knowledge of two communicators have a

similar structure, and the speaker have means to infer this, the speaker will refer

to only a small subset of his or her spatial knowledge of high prominence. Consider

the hierarchically organized spatial knowledge of the speaker as S and the spatial

knowledge of the hearer as H. This prominent knowledge of the speaker and

the hearer are denoted as Sprom and Hprom, respectively. The speaker selects a

referent r, such that r ∈ Sprom, and composes the utterance. The hearer relates

r to her prominent knowledge Hprom, and interprets r, i.e. retrieves the meaning

of r in the given context.

Imagine that r /∈ Hprom. If there is common knowledge of r, but it is not

considered a prominent reference by the hearer (r ∈ H), a reference to such

knowledge requires high cognitive effort of the hearer during interpretation. In

real communication situations, if the interpretation requires excessive cognitive

effort, the hearer seeks confirmation of the interpretation of r or additional sup-

porting information.

Due to the distribution of prominence in experiential hierarchies following

power laws, a relatively low number of trips through the street network should

provide a relatively good coverage of the knowledge for successful communication

131


at least at a coarse level of granularity. Due to distribution of prominence values

in the experiential hierarchies, highly prominent streets have values higher by

magnitudes than those of low prominence. From everyday experience, this is the

case in the majority of occasions. Also note that a certain degree of difference in

individual hierarchies is not a barrier to reaching understanding. The match be-

tween Sprom and Hprom is high when most r ∈ Sprom are also part of Hprom. This

match does not require the same ranking of r in the two respective experiential

hierarchies.

This allows an experiential hierarchical dataset to be constructed, usable by

a navigation service without rich background knowledge of the spatial knowledge

of the user. Individual experiential hierarchies are continuous rankings, and it

is difficult to draw a line separating prominent and non-prominent streets. It

is, however, possible to approximate this limit by the mean value in the distri-

bution. The bulk of the streets in the hierarchy are below the mean value of

prominence. As shown earlier, the model of destination descriptions presented

allows for alteration of this threshold as the only customization option. Further-

more, the integration of hierarchies of different types of elements of the city allows

for substitution of a different referent when, e.g., no prominent path is available

as reference. Should automated systems with inferential interfaces be built, the

interface should provide the user with means to provide feedback.

7.3.3 Experiential Urban Data Structures for Destination

Descriptions

The concept of construction of integrated datasets presented, revealing the expe-

riential hierarchy of the urban structure, is based on a predominantly structuralist

approach to the quantification of the perception of prominence of the elements

organized in the hierarchy. Structural properties of the elements of the city,

along with their visual properties were assessed to produce estimates of promi-

nence. The resulting hierarchical structures allow the common knowledge of the

individual spatial features of the environment studied to be inferred.

The quality of the dataset influences the estimate of relevance of the retrieved

referents. While cognitively plausible, the construction of landmark hierarchies

based exclusively on visibility produced distortions in the dataset, compared to

the perception of the centre of Hannover by locals. Although more appropri-

ate referents could be selected by a local expert, the references selected by the

computational model were judged usable and satisfactory.

The argumentation presented in this thesis starts from the position that

132

M. TOMKO 7.4. OUTLOOK

structural and visual properties of the elements of the city are paramount in their

assessment of prominence. The experience of visual and structural prominence

is common among the population with similar spatial behaviour. The shared

experience is further strengthened by secondary experience of these prominent

features, through indirect sources such as maps, news articles and Web resources.

The inference of semantic prominence of an element, on the other hand,

is problematic and highly subjective (i.e., individual to a person). Semantic

properties of spatial elements can, however, be used in destination descriptions in

an indirect manner. A hearer can be sensitized, or primed, to a specific semantic

characteristic of a landmark by the speaker or the navigation system. Once seen,

the landmark will be perceived as salient. This mechanism allows hearers to

successfully use destination descriptions containing references which may not be

usually perceived as prominent by the hearers.

7.4 Outlook

This thesis introduces a first approach to the concept of destination descriptions,

along with a model enabling their construction. This section collects research

topics related to the modelling of destination descriptions that should be explored

in the future.

7.4.1 Integrating Destination Descriptions with Turn-Based

Directions

A combination of destination descriptions and turn-based route directions is a

common feature of human-generated navigation instructions for people assumed

to have a-priori spatial knowledge of the environment. Destination descriptions

provided at coarser granularities are coupled with added detail in form of turn-

based route directions in the proximity of the destination. The change of the mode

of communication of the spatial knowledge is based on the speaker’s assumption

that the hearer’s spatial knowledge is not complete enough to be able to identify

and reach the destination without the added detail.

The transition to turn-based directions may occur at different distance from

the destination, and need not be direct. Chunking approaches considering the

structural properties of the route can be combined at the finer levels of detail to

further improve the cognitive ergonomics of the resulting directions. Namely, the

Higher-Order Direction Elements (HORDE) of Klippel and co-workers (Klippel,

2003b; Klippel et al., 2003; Richter, 2007a) are of relevance. Based on the purely

133


structural properties of the route, such HORDE complement well the more general

referents provided by the model presented. Furthermore, Richter (2007b) explores

the construction of so-called overview route directions by taking the bottom-up

perspective and starting with the structure of the route itself, extending the work

of Richter (2007a); Richter and Klippel (2005). The combination of destination

descriptions with HORDE-based or overview route directions presents a transition

from a purely hierarchical approach to a turn-based perspective on direction,

restricting the wayfinder to a specific route (Figure 7.1). Recent work of Srinivas

and Hirtle (to appear, 2007) attempts to provide a conceptual analysis of the

content of such route directions from the point of view of occurrence of familiar

and novel knowledge.

Figure 7.1: Destination descriptions transiting into turn-based route directionsin the proximity of the destination.

The transition from destination descriptions to turn-based directions is of

importance especially in less structured parts of the city with low density of

landmarks and large districts. There, more references to paths connecting the

skeleton of prominent paths to the destination are needed. The output of a route-

planning service supporting the recently proposed Cognitive OpenLS standard

integrating the concept of HORDE (Hansen et al., 2006) could conveniently be

coupled with the model of destination descriptions presented to provide a smooth

transition between destination descriptions and turn-based directions.

134

M. TOMKO 7.4. OUTLOOK

7.4.2 The Where? Question

The model of destination descriptions proposed partially builds on previous re-

search on hierarchical place descriptions in large-scale environments. Table-top

scene settings, place descriptions in buildings, or descriptions of text locations in

books (Plumert et al., 1995a, 2001; Shanon, 1979) have previously been described

and analysed. The model proposed provides destination descriptions in space of

environmental scale (Montello, 1993), thus not perceivable from a single point.

The verification of the possibility of the application of the principles of selection

of references for destination descriptions in a large-scale setting is a question for

further research.

If the model could be generalized across a multitude of spatial scales, the

characteristics of the results could be generalized as the properties of answers to

generic Where? questions. Automated scene description systems could then be

developed, enabling, e.g., to communicate spatial descriptions to people through

voice-based interfaces.

7.4.3 Externalization of Destination Descriptions

The model proposed focuses on the selection of relevant references for inclusion in

destination descriptions. A navigation service providing destination descriptions

following the model proposed will need to externalize the information in a form

adapted to the users of the system. Externalization methods, such as natural

language generation or schematic visualization interfaces need to be devised to

communicate the references selected in an appropriate manner.

Furthermore, the user may not be, for various reasons, satisfied by the

model’s selection of references. The information provided may be judged excessive

or insufficient. While people are good at handling omissions and imperfections in

route directions (MacMahon, 2005), well designed user interfaces should cater for

such situations. The visualization of the references provided in destination de-

scriptions is possible through hierarchical collapsing structures, enabling the dis-

closure of more or less information for a specific part of the route. Such interfaces

provide a user with means to transit between destination descriptions and turn-

based direction at will. Dialog based systems (see, e.g. www.talk-project.org)

may provide such possibilities in speech-based interfaces.

135


7.4.4 Coupling of Inferential and Agent-Based Systems

Customization of route directions for users familiar with the environment was,

until now, a largely neglected area of research. The inferential model of selection

of references for destination descriptions provides an alternative solution to per-

sonalization of route directions through agent-based systems. Such systems relied

on storing personal profiles of users to adapt the content of future route direction

requests to the preferences of the users, such as speed vs. length of the route

(Rogers et al., 1999; Wagner et al., 2002). Agent based systems rely on a learn-

ing software embedded in personal mobile devices, such as mobile phones. These

agents sense the actions of the users as they navigate in a given environment,

store, and use this historic information for optimized, personalized information

provision in the future (Patel et al., 2006).

When first initialized, the agent has no previous knowledge of the user’s

knowledge (unless the user explicitly declares some knowledge in their profile),

and thus the information retrieved is not personalized. As the user continues

to use the device with the software agent tracking their (spatial) behaviour, the

agent learns more about the user and is able to use this information to provide

more and more customized information, such as route directions with familiar

landmarks. Ultimately, the information provided is fully customized. At this

stage, if a different user would use the device with the mobile agent, the infor-

mation provided may not be customized to his or her needs at all, to the extent

of possibly being insufficient or even misleading.

On the other hand, the model presented infers the relevance of references

to spatial features without prior explicit personalization. The inferential model

proposed provides information customized to a wider audience from the very

first use. Thus, if compared with agent-based systems, the results provided by

the model of destination descriptions presented will be better adapted for the

information needs of the local from the very beginning of its usage. The relative

difference between the two approaches will decrease with usage and ultimately

the agent-based system will be able to provide more relevant, better adapted

information to its user than the model of destination descriptions presented in

this thesis. Ultimately, the combination with the mentioned mobile agent-based

systems would provide a fully adaptive system.

7.4.5 Complex Integrated Experiential Hierarchies

Further work is necessary to extend the concept of integrated experiential hierar-

chies to cater for references to nodes and barriers, including prominent complex

136

M. TOMKO 7.5. CONCLUDING SCENARIO

configurations of multiple spatial elements of different types. It is hypothesized

that experiential hierarchies of nodes may be constructed using network analysis

approaches similarly to path experiential hierarchies.

Prior to constructing experiential hierarchies of barriers, a thorough defini-

tion of barriers will have to be attempted. Barriers structurally divide the urban

environment, and their prominence is therefore relative to the prominence of the

elements on both sides of the barrier. An automated identification of barriers in

the urban environment may be attempted by first exploring the structural rela-

tion between paths and districts. Districts could be redefined as cohesive parts of

the path network. The identification of barriers as breaks between cohesive parts

of the network is then plausible. Another possibility is to model the prominence

of barriers as the function of the prominence of the segregated districts. Research

in this direction is currently ongoing.

7.5 Concluding Scenario

The morning after her business trip, Stephanie wakes up and, while eating break-

fast, reads her emails. A friend is curious about how her overseas business trip

went and invites Stephanie for lunch to a new cafe in the city. The email in-

cludes the address of the cafe. As Stephanie does not know the place, she selects

the address with the cursor and press a hot-key combination querying her on-line

navigation service of choice. As it happens, her preferred system is just testing

a novel algorithm generating destination descriptions. The system returns a de-

scription of the location of the cafe satisfying Stephanie’s information needs, as

she has been living in the city for the past few years and is familiar with its lay-

out. The destination description is brief and easy to remember, eliminating the

necessity to write down the navigation information on a piece of paper. Stephanie

also appreciates how the system protects her privacy, by not requiring her to enter

her full address to generate route directions. The system only detects the suburb

from which Stephanie connects to the Internet service provider.

Stephanie sets on her way to the cafe. An accident changes the traffic con-

ditions and requires Stephanie to take a detour. Stephanie is confident of finding

an alternative route on her own, even managing to avoid the most congested

streets. The destination descriptions provided by her navigation system remain

usable even though the traffic conditions changed. She reaches the cafe just in

time. And to top it off, the food is excellent!

137


138

Appendix A

Landmark Names

Table A.1: Common names of the landmarks of central Hannover. The names oftheir most prominent reference regions are in parenthesis, where applicable.

Landmark ID Landmark Name

H05V49E Bar at the Maschteich

H06Y0NB Rathaus (Hannover Mitte, Sudstadt)

H09DX6M Kindertagesstatte

H097TLK Universitat Hannover – Welfenschloss (Hannover Nordstadt)

H06KOBS Geodtisches Institut, Universitt Hannover

H05T3WR Wilhelm-Busch-Museum Georgenpalais

H063YJC Universitatsbibliothek

H05RN6G WC and Kiosk at the Herrenhauser Allee

H05TWH4 Parkhaus

H05JJHO Univ. Hannover Institut

H05IWO5 Mensa

H05V43Q Allianz-Hochhaus

H03WTT1 N/A

H03PNBO Polizeidirektion 11

H01F6M0 Marktkirche

H03NG5F Maritim-Hotel (Hannover Mitte)

H04SBR1 Sparkasse Bank

H03PO1O Anzeiger - Hochhaus

H03WUC7 Kreuzkirche

H03PO3Z Univ. Hannover Institut, Chimu Restaurant

H074YH2 Nds. Landesbibliothek

H03P85A Christuskirche

H05MF0G Institute of Chemistry

H03Q6S0 Postverwaltung

H01BHXG ALLBANK

H05V4AW Postamt

H03NGBE Volkshochschule

H04PTS0 Katasteramt (Kropcke)

H06E9I7 Hauptbahnhof

H01P2HN Bratwurst Glockle

H01FM8E Nds. Staatstheater–Oper

H01O23Z Parking garage

139

APPENDIX A. LANDMARK NAMES

140

Appendix B

Input Dataset of Hannover

Module Data of the Haskell implementation of the model of destination descrip-tions where data types are declared and the data input is made.

module Data where

type ObjectName = String --name of all objects - paths , landmarks , districts

type Level = Double -- Int -- for paths , this value has to be computed

type Expvalue = Double -- value of experiential rank Ei of paths

type ObjectID = String -- ID of an Object

type Landmarks = [ObjectID]

type Neighbors = [ObjectID]

type Districts = [ObjectID]

type Path = (ObjectID ,Districts ,Expvalue)

type District = (ObjectID ,Level ,ObjectName ,Landmarks ,Neighbors)

data Object = Area ObjectID Level ObjectName Landmarks Neighbors

| Street ObjectID Districts Expvalue deriving (Show ,Eq)

eimean :: Expvalue -- Mean value of experiential street ranking Ei for Hannover

eimean = 0.001915632

-- TEST DATA level 7 to 2, 53 elements , based on 32 individual districts , 394

paths

--Definition of variable names of type District

h7_H06Y0NB ,h6_H06Y0NB ,h5_H06Y0NB ,h4_H06Y0NB ,h3_H06Y0NB ,h2_H06Y0NB ,h6_H03NG5F ,

h5_H03NG5F ,h4_H03NG5F ,h3_H03NG5F ,h2_H03NG5F ,h5_H097TLK ,h4_H097TLK ,h3_H097TLK ,

h2_H097TLK ,h4_H01F6M0 ,h3_H01F6M0 ,h2_H01F6M0 ,h3_H05T3WR ,h2_H05T3WR ,h3_H06KOBS ,

h2_H06KOBS ,h3_H09DX6M ,h2_H09DX6M ,h3_H05V43Q ,h2_H05V43Q ,h3_H03PO1O ,h2_H03PO1O ,

h3_H063YJC ,h2_H063YJC ,h3_H04PTS0 ,h2_H04PTS0 ,h2_H05V49E ,h2_H05RN6G ,h2_H05TWH4 ,

h2_H05JJHO ,h2_H05IWO5 ,h2_H03WTT1 ,h2_H03PNBO ,h2_H04SBR1 ,h2_H03WUC7 ,h2_H03PO3Z ,

h2_H074YH2 ,h2_H03P85A ,h2_H05MF0G ,h2_H03Q6S0 ,h2_H01BHXG ,h2_H05V4AW ,h2_H03NGBE ,

h2_H06E9I7 ,h2_H01P2HN ,h2_H01FM8E ,h2_H01O23Z :: District

--Definition of variable names of type Path

n20248P ,n01FVOB ,n01FVU1 ,n01FUI0 ,n01FUHY ,n01FUSH ,n20255V ,n01FUP2 ,n202575 ,n01FUPR ,

n01FVOV ,n01FUKW ,n01FUPN ,n01HX6C ,n01H2AO ,n01FUUN ,n01H2IZ ,n202564 ,n01FUFF ,n01FUOQ ,

n01FVP7 ,n01FUFM ,n20255W ,n01FVTJ ,n01H2IU ,n01FUPK ,n20DDG6 ,n01FVU7 ,n01FVVR ,n01FUBD ,

n01H2CR ,n01FVTV ,n01FUJN ,n01H2IY ,n202577 ,n01H2AW ,n01H2RI ,n01FUI5 ,n01H2CI ,n01FUKX ,

n01H2CT ,n01FW21 ,n01FUOH ,n01FVVY ,n01FUTN ,n202569 ,n01FUI3 ,n20B4C7 ,n202561 ,n01FVOP ,

n20GR7U ,n01FUT7 ,n20248Q ,n01FUOX ,n20DDHG ,n01FUT2 ,n202579 ,n01FUSV ,n20256X ,n01FUO7 ,

n01FVYK ,n01FUL6 ,n01FVPP ,n01FUUG ,n01FULD ,n207SYO ,n01FUUA ,n20256O ,n01H2TY ,n01H2G7 ,

n01FVP5 ,n01FUVG ,n20248N ,n01FUJV ,n01FVUC ,n2025UZ ,n01FVOX ,n01H2TL ,n01FUP4 ,n01FUQM ,

n01H2TX ,n01H2Q2 ,n01FUKU ,n20C8MD ,n01FUFK ,n20C8Q0 ,n01FVPX ,n01FVUB ,n01FUUD ,n01FUPO ,

n01FUT5 ,n20C8N3 ,n01FUQO ,n20C8KV ,n01FVUA ,n01FUOE ,n01FUSC ,n01FUJS ,n01FUB0 ,n20DDHX ,

n01FUEC ,n01FVYL ,n20255Y ,n20DDB4 ,n01H2U5 ,n01FVP3 ,n01FVVP ,n01FUKI ,n01FVTY ,n01FVR4 ,

n01H2R6 ,n20DDIT ,n01FUV6 ,n01H2PP ,n01FVYX ,n01FU95 ,n01FUVF ,n01FUSM ,n01FVQS ,n20255X ,

n01FUU3 ,n01H2GJ ,n01H2Q3 ,n01FUOK ,n20C8N2 ,n01FW2N ,n01H2RE ,n01FVRO ,n01FUGX ,n01FVO5 ,

n01FUVH ,n01FUKF ,n01FUHZ ,n01FUOW ,n01FUPQ ,n01FVO3 ,n01FUTE ,n01FUTF ,n20C8LF ,n01FUV8 ,

n01FVOQ ,n01FVPV ,n01FUOR ,n01FVPM ,n01FVOF ,n01FUUL ,n01FVTS ,n01FUQ6 ,n01FUGV ,n202578 ,

n20256N ,n01FVOH ,n01FUBN ,n01FVRV ,n01FVVX ,n01FVVL ,n01FVUW ,n01FVPN ,n01FUTO ,n20249J ,

141

APPENDIX B. INPUT DATASET OF HANNOVER

n01FVLJ ,n01FVM0 ,n01FVV2 ,n01FUT6 ,n01H2IX ,n01FVQH ,n207T0T ,n20257C ,n2025VG ,n20C8RN ,

n20256R ,n01FULB ,n01FVOR ,n01FULQ ,n01FVP1 ,n01FVO6 ,n01H2PL ,n01FVTK ,n01FVPQ ,n01FUH0 ,

n01FUJX ,n20256M ,n01FUJR ,n01FVPR ,n01FUOF ,n01FVO8 ,n01FVYZ ,n01H2X0 ,n01FUME ,n01FVQC ,

n20GR5B ,n01FVYM ,n01FVU2 ,n01FVTQ ,n01FUPS ,n01H2FY ,n01FUEE ,n20DDIQ ,n01H2RB ,n20GR5C ,

n01FUOC ,n00E2TA ,n00E643 ,n00E64D ,n00E64G ,n00E66G ,n00E66J ,n00E66L ,n01788P ,n01789A ,

n01789B ,n01789D ,n01789E ,n01789F ,n01789L ,n01789R ,n01789S ,n01789T ,n01789U ,n01789V ,

n01789W ,n01789Y ,n01789Z ,n0178A0 ,n0178A1 ,n0178A2 ,n0178A3 ,n0178A4 ,n0178A5 ,n0178A7 ,

0178A8 ,n0178AA ,n01AB6A ,n01AB6B ,n01AB6C ,n01AB6D ,n01AB6E ,n01AB6G ,n01AB6M ,n01AB6N ,

n01AB6O ,n01AB6R ,n01AB6S ,n01ABD8 ,n01ABDD ,n01ABDF ,n01ABDG ,n01ABDH ,n01ABDI ,n01ABDJ ,

n01ABDK ,n01ABDL ,n01ABDM ,n01ABDN ,n01ABDO ,n01ABDP ,n01ABDQ ,n01ABDS ,n01ABDT ,n01ABDW ,

n01ABDZ ,n01ABE0 ,n01ABE1 ,n01ABE4 ,n01ABE5 ,n01ABE7 ,n01ABE9 ,n01ABED ,n01ABEO ,n01ABES ,

n01ABET ,n01ABEV ,n01ABEW ,n01ABEX ,n01ABEY ,n01ABF0 ,n01ABF1 ,n01ABF3 ,n01ABF4 ,n01ABJ8 ,

n01ABJJ ,n01ABJK ,n01ABJX ,n01ABJY ,n01ABJZ ,n01ABK2 ,n01ABK3 ,n01ABK4 ,n01ABKC ,n01ABKF ,

n01ABKG ,n01ABKI ,n01ABKN ,n01ABKO ,n01ABL3 ,n01ABL4 ,n01ABL5 ,n01ABL6 ,n01ABL7 ,n01ABLE ,

n01ABLF ,n01ABOX ,n01ABP6 ,n01ABPC ,n01ABPD ,n01ABPM ,n01ABPP ,n01ABPR ,n01ABPU ,n01ABPW ,

n01ABPZ ,n01ABQ0 ,n01ABQ2 ,n01ABQ3 ,n01ABQ6 ,n01ABQ8 ,n01ABQ9 ,n01ABQB ,n01ABQE ,n01ABQF ,

n01ABQG ,n01ABQI ,n01ABQJ ,n01ABQL ,n01ABQP ,n01ABQQ ,n01ABQS ,n01ABQY ,n01ABQZ ,n01ABR0 ,

n01ABR1 ,n01ABV1 ,n01ABV2 ,n01ABV3 ,n01ABV4 ,n01ABV5 ,n01ABVA ,n01ABVU ,n01ABVV ,n01AC10 ,

n01ADGP ,n01ADH6 ,n01ADHA ,n01ADHB ,n01ADHC ,n01ADHD ,n01ADHE ,n01ADHH ,n01FUL8 ,n01FUSL ,

n01FUTR ,n01FUV1 ,n01FUVA ,n202566 ,n202568 ,n20256D ,n20256P ,n20256S ,n20256U ,n207SXP ,

n207SXQ ,n207SXS ,n207T02 ,n207T5T ,n207T5V ,n207T5W ,n207T5X ,n20B4C4 ,n20C9F0 ,n20DDFW ,

n20DDIH ,n20GR4W ,n01FU8L ,n01FUCA ,n202562 ,n01HX6G ,n01FULP ,n01FUT8 ,n01FVRI ,n20GR51 ,

n01FVZA ,n01FUE6 ,n01FUFE ,n01FVQI ,n01FVR6 ,n01FVZ8 ,n01FUGU ,n01FVQ5 ,n01FUAZ ,n01FVZ6 ,

n01FUP9 ,n01FVPZ ,n01FULE ,n01H2PS :: Path

-- Definition of the list of Districts

areas :: [District]

areas = [h7_H06Y0NB ,h6_H06Y0NB ,h5_H06Y0NB ,h4_H06Y0NB ,h3_H06Y0NB ,h2_H06Y0NB ,

h6_H03NG5F ,h5_H03NG5F ,h4_H03NG5F ,h3_H03NG5F ,h2_H03NG5F ,h5_H097TLK ,h4_H097TLK ,

h3_H097TLK ,h2_H097TLK ,h4_H01F6M0 ,h3_H01F6M0 ,h2_H01F6M0 ,h3_H05T3WR ,h2_H05T3WR ,

h3_H06KOBS ,h2_H06KOBS ,h3_H09DX6M ,h2_H09DX6M ,h3_H05V43Q ,h2_H05V43Q ,h3_H03PO1O ,

h2_H03PO1O ,h3_H063YJC ,h2_H063YJC ,h3_H04PTS0 ,h2_H04PTS0 ,h2_H05V49E ,h2_H05RN6G ,

h2_H05TWH4 ,h2_H05JJHO ,h2_H05IWO5 ,h2_H03WTT1 ,h2_H03PNBO ,h2_H04SBR1 ,h2_H03WUC7 ,

h2_H03PO3Z ,h2_H074YH2 ,h2_H03P85A ,h2_H05MF0G ,h2_H03Q6S0 ,h2_H01BHXG ,h2_H05V4AW ,

h2_H03NGBE ,h2_H06E9I7 ,h2_H01P2HN ,h2_H01FM8E ,h2_H01O23Z]

-- Definition of the list of Paths

paths :: [Path]

paths =

[n20248P ,n01FVOB ,n01FVU1 ,n01FUI0 ,n01FUHY ,n01FUSH ,n20255V ,n01FUP2 ,n202575 ,

n01FUPR ,n01FVOV ,n01FUKW ,n01FUPN ,n01HX6C ,n01H2AO ,n01FUUN ,n01H2IZ ,n202564 ,n01FUFF ,

n01FUOQ ,n01FVP7 ,n01FUFM ,n20255W ,n01FVTJ ,n01H2IU ,n01FUPK ,n20DDG6 ,n01FVU7 ,n01FVVR ,

n01FUBD ,n01H2CR ,n01FVTV ,n01FUJN ,n01H2IY ,n202577 ,n01H2AW ,n01H2RI ,n01FUI5 ,n01H2CI ,

n01FUKX ,n01H2CT ,n01FW21 ,n01FUOH ,n01FVVY ,n01FUTN ,n202569 ,n01FUI3 ,n20B4C7 ,n202561 ,

n01FVOP ,n20GR7U ,n01FUT7 ,n20248Q ,n01FUOX ,n20DDHG ,n01FUT2 ,n202579 ,n01FUSV ,n20256X ,

n01FUO7 ,n01FVYK ,n01FUL6 ,n01FVPP ,n01FUUG ,n01FULD ,n207SYO ,n01FUUA ,n20256O ,n01H2TY ,

n01H2G7 ,n01FVP5 ,n01FUVG ,n20248N ,n01FUJV ,n01FVUC ,n2025UZ ,n01FVOX ,n01H2TL ,n01FUP4 ,

n01FUQM ,n01H2TX ,n01H2Q2 ,n01FUKU ,n20C8MD ,n01FUFK ,n20C8Q0 ,n01FVPX ,n01FVUB ,n01FUUD ,

n01FUPO ,n01FUT5 ,n20C8N3 ,n01FUQO ,n20C8KV ,n01FVUA ,n01FUOE ,n01FUSC ,n01FUJS ,n01FUB0 ,

n20DDHX ,n01FUEC ,n01FVYL ,n20255Y ,n20DDB4 ,n01H2U5 ,n01FVP3 ,n01FVVP ,n01FUKI ,n01FVTY ,

n01FVR4 ,n01H2R6 ,n20DDIT ,n01FUV6 ,n01H2PP ,n01FVYX ,n01FU95 ,n01FUVF ,n01FUSM ,n01FVQS ,

n20255X ,n01FUU3 ,n01H2GJ ,n01H2Q3 ,n01FUOK ,n20C8N2 ,n01FW2N ,n01H2RE ,n01FVRO ,n01FUGX ,

n01FVO5 ,n01FUVH ,n01FUKF ,n01FUHZ ,n01FUOW ,n01FUPQ ,n01FVO3 ,n01FUTE ,n01FUTF ,n20C8LF ,

n01FUV8 ,n01FVOQ ,n01FVPV ,n01FUOR ,n01FVPM ,n01FVOF ,n01FUUL ,n01FVTS ,n01FUQ6 ,n01FUGV ,

n202578 ,n20256N ,n01FVOH ,n01FUBN ,n01FVRV ,n01FVVX ,n01FVVL ,n01FVUW ,n01FVPN ,n01FUTO ,

n20249J ,n01FVLJ ,n01FVM0 ,n01FVV2 ,n01FUT6 ,n01H2IX ,n01FVQH ,n207T0T ,n20257C ,n2025VG ,

n20C8RN ,n20256R ,n01FULB ,n01FVOR ,n01FULQ ,n01FVP1 ,n01FVO6 ,n01H2PL ,n01FVTK ,n01FVPQ ,

n01FUH0 ,n01FUJX ,n20256M ,n01FUJR ,n01FVPR ,n01FUOF ,n01FVO8 ,n01FVYZ ,n01H2X0 ,n01FUME ,

n01FVQC ,n20GR5B ,n01FVYM ,n01FVU2 ,n01FVTQ ,n01FUPS ,n01H2FY ,n01FUEE ,n20DDIQ ,n01H2RB ,

n20GR5C ,n01FUOC ,n00E2TA ,n00E643 ,n00E64D ,n00E64G ,n00E66G ,n00E66J ,n00E66L ,n01788P ,

n01789A ,n01789B ,n01789D ,n01789E ,n01789F ,n01789L ,n01789R ,n01789S ,n01789T ,n01789U ,

n01789V ,n01789W ,n01789Y ,n01789Z ,n0178A0 ,n0178A1 ,n0178A2 ,n0178A3 ,n0178A4 ,n0178A5 ,

n0178A7 ,n0178A8 ,n0178AA ,n01AB6A ,n01AB6B ,n01AB6C ,n01AB6D ,n01AB6E ,n01AB6G ,n01AB6M ,

n01AB6N ,n01AB6O ,n01AB6R ,n01AB6S ,n01ABD8 ,n01ABDD ,n01ABDF ,n01ABDG ,n01ABDH ,n01ABDI ,

n01ABDJ ,n01ABDK ,n01ABDL ,n01ABDM ,n01ABDN ,n01ABDO ,n01ABDP ,n01ABDQ ,n01ABDS ,n01ABDT ,

n01ABDW ,n01ABDZ ,n01ABE0 ,n01ABE1 ,n01ABE4 ,n01ABE5 ,n01ABE7 ,n01ABE9 ,n01ABED ,n01ABEO ,

n01ABES ,n01ABET ,n01ABEV ,n01ABEW ,n01ABEX ,n01ABEY ,n01ABF0 ,n01ABF1 ,n01ABF3 ,n01ABF4 ,

n01ABJ8 ,n01ABJJ ,n01ABJK ,n01ABJX ,n01ABJY ,n01ABJZ ,n01ABK2 ,n01ABK3 ,n01ABK4 ,n01ABKC ,

n01ABKF ,n01ABKG ,n01ABKI ,n01ABKN ,n01ABKO ,n01ABL3 ,n01ABL4 ,n01ABL5 ,n01ABL6 ,n01ABL7 ,

n01ABLE ,n01ABLF ,n01ABOX ,n01ABP6 ,n01ABPC ,n01ABPD ,n01ABPM ,n01ABPP ,n01ABPR ,n01ABPU ,

n01ABPW ,n01ABPZ ,n01ABQ0 ,n01ABQ2 ,n01ABQ3 ,n01ABQ6 ,n01ABQ8 ,n01ABQ9 ,n01ABQB ,n01ABQE ,

n01ABQF ,n01ABQG ,n01ABQI ,n01ABQJ ,n01ABQL ,n01ABQP ,n01ABQQ ,n01ABQS ,n01ABQY ,n01ABQZ ,

n01ABR0 ,n01ABR1 ,n01ABV1 ,n01ABV2 ,n01ABV3 ,n01ABV4 ,n01ABV5 ,n01ABVA ,n01ABVU ,n01ABVV ,

n01AC10 ,n01ADGP ,n01ADH6 ,n01ADHA ,n01ADHB ,n01ADHC ,n01ADHD ,n01ADHE ,n01ADHH ,n01FUL8 ,

142

M. TOMKO

n01FUSL ,n01FUTR ,n01FUV1 ,n01FUVA ,n202566 ,n202568 ,n20256D ,n20256P ,n20256S ,n20256U ,

n207SXP ,n207SXQ ,n207SXS ,n207T02 ,n207T5T ,n207T5V ,n207T5W ,n207T5X ,n20B4C4 ,n20C9F0 ,

n20DDFW ,n20DDIH ,n20GR4W ,n01FU8L ,n01FUCA ,n202562 ,n01HX6G ,n01FULP ,n01FUT8 ,n01FVRI ,

n20GR51 ,n01FVZA ,n01FUE6 ,n01FUFE ,n01FVQI ,n01FVR6 ,n01FVZ8 ,n01FUGU ,n01FVQ5 ,n01FUAZ ,

n01FVZ6 ,n01FUP9 ,n01FVPZ ,n01FULE ,n01H2PS]

-- Data entries for the instances of Districts at consecutive granularities

h7_H06Y0NB = ("h7_H06Y0NB", 7, "H06Y0NB", ["h6_H06Y0NB", "h6_H03NG5F"], [])

h6_H06Y0NB = ("h6_H06Y0NB", 6, "H06Y0NB", ["h5_H06Y0NB", "h5_H03NG5F",

"h5_H097TLK"], ["h6_H03NG5F"])


"h4_H01F6M0"], ["h5_H03NG5F"])


"h3_H01F6M0", "h3_H04PTS0"], ["h4_H03NG5F"])

h3_H06Y0NB = ("h3_H06Y0NB", 3, "H06Y0NB", ["h2_H05V49E", "h2_H06Y0NB",

"h2_H03PNBO", "h2_H03NG5F", "h2_H03NGBE", "h2_H03WTT1", "h2_H01F6M0",

"h2_H05V4AW"], ["h3_H03NG5F"])

h2_H06Y0NB = ("h2_H06Y0NB", 2, "H06Y0NB", [], ["h2_H05V49E", "h2_H03PNBO",

"h2_H03NG5F", "h2_H03NGBE"])

h6_H03NG5F = ("h6_H03NG5F", 6, "H03NG5F", ["h5_H097TLK", "h5_H03NG5F"],

["h6_H06Y0NB"])

h5_H03NG5F = ("h5_H03NG5F", 5, "H03NG5F", ["h4_H097TLK", "h4_H01F6M0",

"h4_H03NG5F"], ["h5_H06Y0NB", "h5_H097TLK"])

h4_H03NG5F = ("h4_H03NG5F", 4, "H03NG5F", ["h3_H05V43Q", "h3_H01F6M0",

"h3_H03NG5F", "h3_H03PO1O", "h3_H04PTS0"], ["h4_H06Y0NB", "h4_H01F6M0"])

h3_H03NG5F = ("h3_H03NG5F", 3, "H03NG5F", ["h2_H03WTT1", "h2_H03PNBO",

"h2_H01F6M0", "h2_H03NG5F", "h2_H03WUC7", "h2_H05V4AW", "h2_H04PTS0",

"h2_H01P2HN", "h2_H03NGBE", "h2_H01FM8E"], ["h3_H06Y0NB", "h3_H01F6M0",

"h3_H04PTS0"])

h2_H03NG5F = ("h2_H03NG5F", 2, "H03NG5F", [], ["h2_H06Y0NB", "h2_H03PNBO",

"h2_H01F6M0", "h2_H05V4AW", "h2_H03NGBE"])

h5_H097TLK = ("h5_H097TLK", 5, "H097TLK", ["h4_H097TLK", "h4_H01F6M0"],

["h5_H097TLK"])

h4_H097TLK = ("h4_H097TLK", 4, "H097TLK", ["h3_H097TLK", "h3_H05T3WR",

"h3_H06KOBS", "h3_H09DX6M", "h3_H063YJC", "h3_H05V43Q"], ["h4_H01F6M0"])

h3_H097TLK = ("h3_H097TLK", 3, "H097TLK", ["h2_H09DX6M", "h2_H097TLK",

"h2_H06KOBS", "h2_H05IWO5", "h2_H05MF0G", "h2_H05T3WR", "h2_H063YJC",

"h2_H05JJHO", "h2_H05RN6G", "h2_H05TWH4", "h2_H05V43Q", "h2_H074YH2",

"h2_H03P85A"], ["h3_H05T3WR", "h3_H06KOBS", "h3_H09DX6M", "h3_H063YJC"])

h2_H097TLK = ("h2_H097TLK", 2, "H097TLK", [], ["h2_H09DX6M", "h2_H06KOBS",

"h2_H05T3WR", "h2_H063YJC", "h2_H05MF0G"])

h4_H01F6M0 = ("h4_H01F6M0", 4, "H01F6M0", ["h3_H05V43Q", "h3_H01F6M0",

"h3_H03PO1O", "h3_H04PTS0", "h3_H063YJC"], ["h4_H097TLK", "h4_H03NG5F"])

h3_H01F6M0 = ("h3_H01F6M0", 3, "H01F6M0", ["h2_H05V43Q", "h2_H03WTT1",

"h2_H03PNBO", "h2_H01F6M0", "h2_H04SBR1", "h2_H03WUC7", "h2_H01BHXG",

"h2_H01P2HN", "h2_H05V4AW", "h2_H04PTS0", "h2_H06E9I7", "h2_H01FM8E",

"h2_H01O23Z"], ["h3_H05V43Q", "h3_H03NG5F", "h3_H03PO1O", "h3_H04PTS0"])

h2_H01F6M0 = ("h2_H01F6M0", 2, "H01F6M0", [], ["h2_H03WTT1", "h2_H03PNBO",

"h2_H03NG5F", "h2_H03WUC7", "h2_H05V4AW", "h2_H04PTS0", "h2_H01P2HN"])

h3_H05T3WR = ("h3_H05T3WR", 3, "H05T3WR", ["h2_H05T3WR", "h2_H05RN6G",

"h2_H05TWH4", "h2_H05JJHO"], ["h3_H097TLK", "h3_H06KOBS"])

h2_H05T3WR = ("h2_H05T3WR", 2, "H05T3WR", [], ["h2_H097TLK", "h2_H06KOBS",

"h2_H05RN6G", "h2_H05TWH4", "h2_H05JJHO"])

h3_H06KOBS = ("h3_H06KOBS", 3, "H06KOBS", ["h2_H06KOBS", "h2_H05T3WR",

"h2_H05TWH4", "h2_H05JJHO", "h2_H05IWO5"], ["h3_H097TLK", "h3_H05T3WR",

"h3_H09DX6M"])

h2_H06KOBS = ("h2_H06KOBS", 2, "H06KOBS", [], ["h2_H09DX6M", "h2_H097TLK",

"h2_H05T3WR", "h2_H05JJHO", "h2_H05IWO5"])

h3_H09DX6M = ("h3_H09DX6M", 3, "H09DX6M", ["h2_H09DX6M", "h2_H06KOBS",

"h2_H05JJHO", "h2_H05IWO5", "h2_H05MF0G"], ["h3_H097TLK", "h3_H06KOBS"])

h2_H09DX6M = ("h2_H09DX6M", 2, "H09DX6M", [], ["h2_H097TLK", "h2_H06KOBS",

"h2_H05IWO5", "h2_H05MF0G"])

h3_H05V43Q = ("h3_H05V43Q", 3, "H05V43Q", ["h2_H063YJC", "h2_H05V43Q",

"h2_H03WTT1", "h2_H04SBR1", "h2_H03PO3Z", "h2_H074YH2", "h2_H03PO1O",

"h2_H03WUC7", "h2_H01BHXG", "h2_H03P85A"], ["h3_H01F6M0", "h3_H03PO1O",

"h3_H063YJC"])

143


h2_H05V43Q = ("h2_H05V43Q", 2, "H05V43Q", [], ["h2_H03WTT1", "h2_H04SBR1",

"h2_H03PO3Z", "h2_H074YH2"])

h3_H03PO1O = ("h3_H03PO1O", 3, "H03PO1O", ["h2_H04SBR1", "h2_H03PO1O",

"h2_H03PO3Z", "h2_H03P85A", "h2_H03Q6S0", "h2_H01BHXG", "h2_H06E9I7",

"h2_H01O23Z", "h2_H03WUC7", "h2_H01P2HN "], ["h3_H05V43Q", "h3_H01F6M0",

"h3_H063YJC", "h3_H04PTS0"])

h2_H03PO1O = ("h2_H03PO1O", 2, "H03PO1O", [], ["h2_H04SBR1", "h2_H03PO3Z",

"h2_H03P85A", "h2_H03Q6S0", "h2_H01BHXG"])

h3_H063YJC = ("h3_H063YJC", 3, "H063YJC", ["h2_H063YJC", "h2_H03PO1O",

"h2_H03PO3Z", "h2_H074YH2", "h2_H03P85A", "h2_H05MF0G", "h2_H03Q6S0"],

["h3_H097TLK", "h3_H05V43Q", "h3_H03PO1O"])

h2_H063YJC = ("h2_H063YJC", 2, "H063YJC", [], ["h2_H097TLK", "h2_H074YH2",

"h2_H03P85A", "h2_H05MF0G"])

h3_H04PTS0 = ("h3_H04PTS0", 3, "H04PTS0", ["h2_H03Q6S0", "h2_H04PTS0",

"h2_H06E9I7", "h2_H01FM8E", "h2_H01O23Z", "h2_H05V4AW", "h2_H01BHXG",

"h2_H01P2HN"], ["h3_H01F6M0", "h3_H03NG5F", "h3_H03PO1O"])

h2_H04PTS0 = ("h2_H04PTS0", 2, "H04PTS0", [], ["h2_H01F6M0", "h2_H05V4AW",

"h2_H06E9I7", "h2_H01P2HN", "h2_H01FM8E", "h2_H01O23Z"])

h2_H05V49E = ("h2_H05V49E", 2, "H05V49E", [], ["h2_H06Y0NB", "h2_H03PNBO"])

h2_H05RN6G = ("h2_H05RN6G", 2, "H05RN6G", [], ["h2_H05T3WR", "h2_H05TWH4"])

h2_H05TWH4 = ("h2_H05TWH4", 2, "H05TWH4", [], ["h2_H05T3WR", "h2_H05RN6G",

"h2_H05JJHO"])

h2_H05JJHO = ("h2_H05JJHO", 2, "H05JJHO", [], ["h2_H06KOBS", "h2_H05T3WR",

"h2_H05TWH4", "h2_H05IWO5"])

h2_H05IWO5 = ("h2_H05IWO5", 2, "H05IWO5", [], ["h2_H09DX6M", "h2_H06KOBS",

"h2_H05JJHO", "h2_H05MF0G"])

h2_H03WTT1 = ("h2_H03WTT1", 2, "H03WTT1", [], ["h2_H05V43Q", "h2_H03PNBO",

"h2_H01F6M0", "h2_H04SBR1", "h2_H03WUC7"])

h2_H03PNBO = ("h2_H03PNBO", 2, "H03PNBO", [], ["h2_H05V49E", "h2_H06Y0NB",

"h2_H03WTT1", "h2_H01F6M0", "h2_H03NG5F"])

h2_H04SBR1 = ("h2_H04SBR1", 2, "H04SBR1", [], ["h2_H05V43Q", "h2_H03WTT1",

"h2_H03PO1O", "h2_H03WUC7", "h2_H03PO3Z", "h2_H01BHXG"])

h2_H03WUC7 = ("h2_H03WUC7", 2, "H03WUC7", [], ["h2_H03WTT1", "h2_H01F6M0",

"h2_H04SBR1", "h2_H01BHXG", "h2_H01P2HN"])

h2_H03PO3Z = ("h2_H03PO3Z", 2, "H03PO3Z", [], ["h2_H05V43Q", "h2_H04SBR1",

"h2_H03PO1O", "h2_H074YH2", "h2_H03P85A"])

h2_H074YH2 = ("h2_H074YH2", 2, "H074YH2", [], ["h2_H063YJC", "h2_H05V43Q",

"h2_H03PO3Z", "h2_H03P85A"])

h2_H03P85A = ("h2_H03P85A", 2, "H03P85A", [], ["h2_H063YJC", "h2_H03PO1O",

"h2_H03PO3Z", "h2_H074YH2"])

h2_H05MF0G = ("h2_H05MF0G", 2, "H05MF0G", [], ["h2_H09DX6M", "h2_H097TLK",

"h2_H063YJC", "h2_H05IWO5"])

h2_H03Q6S0 = ("h2_H03Q6S0", 2, "H03Q6S0", [], ["h2_H03PO1O", "h2_H01BHXG",

"h2_H06E9I7", "h2_H01O23Z"])

h2_H01BHXG = ("h2_H01BHXG", 2, "H01BHXG", [], ["h2_H04SBR1", "h2_H03PO1O",

"h2_H03WUC7", "h2_H03Q6S0", "h2_H01P2HN", "h2_H01O23Z"])

h2_H05V4AW = ("h2_H05V4AW", 2, "H05V4AW", [], ["h2_H01F6M0", "h2_H03NG5F",

"h2_H03NGBE", "h2_H04PTS0", "h2_H01FM8E"])

h2_H03NGBE = ("h2_H03NGBE", 2, "H03NGBE", [], ["h2_H06Y0NB", "h2_H03NG5F",

"h2_H05V4AW"])

h2_H06E9I7 = ("h2_H06E9I7", 2, "H06E9I7", [], ["h2_H03Q6S0", "h2_H04PTS0",

"h2_H01FM8E", "h2_H01O23Z"])

h2_H01P2HN = ("h2_H01P2HN", 2, "H01P2HN", [], ["h2_H01F6M0", "h2_H03WUC7",

"h2_H01BHXG", "h2_H04PTS0", "h2_H01O23Z"])

h2_H01FM8E = ("h2_H01FM8E", 2, "H01FM8E", [], ["h2_H05V4AW", "h2_H04PTS0",

"h2_H06E9I7"])

h2_H01O23Z = ("h2_H01O23Z", 2, "H01O23Z", [], ["h2_H03Q6S0", "h2_H01BHXG",

"h2_H04PTS0", "h2_H06E9I7", "h2_H01P2HN"])

-- Data entries for the instances of Paths

n20248P = ("N20248P", ["h2_H06Y0NB", "h2_H05V4AW", "h2_H03NGBE"], 0.00012742)

n01FVOB = ("N01FVOB", ["h2_H03NGBE"], 0.00005190)

n01FVU1 = ("N01FVU1", ["h2_H01FM8E"], 0.00000176)

n01FUI0 = ("N01FUI0", ["h2_H05TWH4"], 0.00003570)

n01FUHY = ("N01FUHY", ["h2_H05TWH4"], 0.00007140)

n01FUSH = ("N01FUSH", ["h2_H03WTT1", "h2_H03PNBO"], 0.00064386)

n20255V = ("N20255V", ["h2_H03WTT1"], 0.00049183)

n01FUP2 = ("N01FUP2", ["h2_H063YJC", "h2_H03P85A"], 0.00020312)

n202575 = ("N202575", ["h2_H03WTT1"], 0.00085250)

144

M. TOMKO

n01FUPR = ("N01FUPR", ["h2_H063YJC", "h2_H05MF0G"], 0.00039174)

n01FVOV = ("N01FVOV", ["h2_H01F6M0"], 0.00007486)

n01FUKW = ("N01FUKW", ["h2_H097TLK"], 0.00013896)

n01FUPN = ("N01FUPN", ["h2_H063YJC"], 0.00010767)

n01HX6C = ("N01HX6C", ["h2_H03PNBO"], 0.00072974)

n01H2AO = ("N01H2AO", ["h2_H04SBR1", "h2_H01BHXG"], 0.00138521)

n01FUUN = ("N01FUUN", ["h2_H03PO3Z", "h2_H074YH2"], 0.00053495)

n01H2IZ = ("N01H2IZ", ["h2_H03PNBO"], 0.00126691)

n202564 = ("N202564", ["h2_H03P85A"], 0.00014345)

n01FUFF = ("N01FUFF", ["h2_H05RN6G"], 0.00000000)

n01FUOQ = ("N01FUOQ", ["h2_H05V43Q"], 0.00009544)

n01FVP7 = ("N01FVP7", ["h2_H04PTS0", "h2_H01O23Z"], 0.00207503)

n01FUFM = ("N01FUFM", ["h2_H05TWH4", "h2_H05MF0G", "h2_H05JJHO", "h2_H05IWO5"],

0.00032825)

n20255W = ("N20255W", ["h2_H03WTT1", "h2_H03PNBO"], 0.00128772)

n01FVTJ = ("N01FVTJ", ["h2_H05V4AW"], 0.00009124)

n01H2IU = ("N01H2IU", ["h2_H05V49E"], 0.00000000)

n01FUPK = ("N01FUPK", ["h2_H05MF0G"], 0.00000960)

n20DDG6 = ("N20DDG6", ["h2_H05V49E"], 0.00000000)

n01FVU7 = ("N01FVU7", ["h2_H06E9I7"], 0.00001490)

n01FVVR = ("N01FVVR", ["h2_H04PTS0", "h2_H06E9I7"], 0.00005116)

n01FUBD = ("N01FUBD", ["h2_H03WUC7"], 0.00012582)

n01H2CR = ("N01H2CR", ["h2_H05V4AW"], 0.00118043)

n01FVTV = ("N01FVTV", ["h2_H03PNBO", "h2_H05V49E"], 0.00015203)

n01FUJN = ("N01FUJN", ["h2_H05RN6G"], 0.00000000)

n01H2IY = ("N01H2IY", ["h2_H03PO3Z", "h2_H05V43Q"], 0.00012986)

n202577 = ("N202577", ["h2_H01FM8E"], 0.00000176)

n01H2AW = ("N01H2AW", ["h2_H06Y0NB"], 0.00000513)

n01H2RI = ("N01H2RI", ["h2_H05TWH4", "h2_H05JJHO"], 0.00007140)

n01FUI5 = ("N01FUI5", ["h2_H01F6M0"], 0.00013211)

n01H2CI = ("N01H2CI", ["h2_H03PNBO"], 0.00015203)

n01FUKX = ("N01FUKX", ["h2_H05V4AW", "h2_H03NGBE"], 0.00045636)

n01H2CT = ("N01H2CT", ["h2_H03PNBO", "h2_H05V49E"], 0.00075001)

n01FW21 = ("N01FW21", ["h2_H03PO1O", "h2_H03Q6S0"], 0.00034766)

n01FUOH = ("N01FUOH", ["h2_H03P85A", "h2_H074YH2"], 0.00147766)

n01FVVY = ("N01FVVY", ["h2_H03WTT1", "h2_H01F6M0", "h2_H03WUC7"], 0.00295238)

n01FUTN = ("N01FUTN", ["h2_H05RN6G"], 0.00000000)

n202569 = ("N202569", ["h2_H03WTT1"], 0.00859061)

n01FUI3 = ("N01FUI3", ["h2_H05TWH4", "h2_H05MF0G", "h2_H05JJHO", "h2_H05IWO5"],

0.00018402)

n20B4C7 = ("N20B4C7", ["h2_H03Q6S0"], 0.00577930)

n202561 = ("N202561", ["h2_H03WTT1"], 0.00095087)

n01FVOP = ("N01FVOP", ["h2_H03WUC7"], 0.00018301)

n20GR7U = ("N20GR7U", ["h2_H06Y0NB", "h2_H03PNBO", "h2_H05V49E", "h2_H03NG5F"],

0.01016487)

n01FUT7 = ("N01FUT7", ["h2_H03WTT1"], 0.00049183)

n20248Q = ("N20248Q", ["h2_H03NGBE"], 0.00003892)

n01FUOX = ("N01FUOX", ["h2_H063YJC"], 0.00010767)

n20DDHG = ("N20DDHG", ["h2_H06E9I7"], 0.00014789)

n01FUT2 = ("N01FUT2", ["h2_H03WTT1"], 0.00085250)

n202579 = ("N202579", ["h2_H06E9I7"], 0.00000798)

n01FUSV = ("N01FUSV", ["h2_H063YJC", "h2_H074YH2"], 0.00060820)

n20256X = ("N20256X", ["h2_H05MF0G"], 0.00000320)

n01FUO7 = ("N01FUO7", ["h2_H06Y0NB"], 0.00000296)

n01FVYK = ("N01FVYK", ["h2_H05T3WR"], 0.00001711)

n01FUL6 = ("N01FUL6", ["h2_H05V4AW"], 0.00120324)

n01FVPP = ("N01FVPP", ["h2_H05V4AW", "h2_H03NGBE"], 0.00024892)

n01FUUG = ("N01FUUG", ["h2_H05V43Q"], 0.00026088)

n01FULD = ("N01FULD", ["h2_H05MF0G", "h2_H05IWO5"], 0.00000320)

n207SYO = ("N207SYO", ["h2_H05V49E"], 0.00000000)

n01FUUA = ("N01FUUA", ["h2_H03WUC7"], 0.00012963)

n20256O = ("N20256O", ["h2_H03PO1O"], 0.00003772)

n01H2TY = ("N01H2TY", ["h2_H04PTS0", "h2_H01O23Z"], 0.00192356)

n01H2G7 = ("N01H2G7", ["h2_H01F6M0", "h2_H03WUC7"], 0.00012325)

n01FVP5 = ("N01FVP5", ["h2_H01F6M0", "h2_H04PTS0"], 0.00009005)

n01FUVG = ("N01FUVG", ["h2_H03P85A"], 0.00034758)

n20248N = ("N20248N", ["h2_H05V49E"], 0.00000000)

n01FUJV = ("N01FUJV", ["h2_H03Q6S0"], 0.00007817)

n01FVUC = ("N01FVUC", ["h2_H05MF0G"], 0.00000363)

n2025UZ = ("N2025UZ", ["h2_H01F6M0"], 0.00007046)

n01FVOX = ("N01FVOX", ["h2_H03PNBO"], 0.00030406)

n01H2TL = ("N01H2TL", ["h2_H05V43Q"], 0.00009544)

n01FUP4 = ("N01FUP4", ["h2_H03PO3Z"], 0.00003442)

n01FUQM = ("N01FUQM", ["h2_H01FM8E"], 0.00000340)

145


n01H2TX = ("N01H2TX", ["h2_H04PTS0", "h2_H01P2HN"], 0.00090820)

n01H2Q2 = ("N01H2Q2", ["h2_H06Y0NB"], 0.00000296)

n01FUKU = ("N01FUKU", ["h2_H05TWH4", "h2_H063YJC", "h2_H097TLK", "h2_H05RN6G",

"h2_H05JJHO", "h2_H06KOBS"], 0.00428415)

n20C8MD = ("N20C8MD", ["h2_H05TWH4"], 0.00051885)

n01FUFK = ("N01FUFK", ["h2_H05TWH4", "h2_H05RN6G"], 0.00504333)

n20C8Q0 = ("N20C8Q0", ["h2_H01O23Z", "h2_H03Q6S0"], 0.00012990)

n01FVPX = ("N01FVPX", ["h2_H01O23Z", "h2_H03Q6S0"], 0.01091186)

n01FVUB = ("N01FVUB", ["h2_H01F6M0"], 0.00006606)

n01FUUD = ("N01FUUD", ["h2_H01F6M0"], 0.00061213)

n01FUPO = ("N01FUPO", ["h2_H063YJC", "h2_H05MF0G", "h2_H097TLK"], 0.00027940)

n01FUT5 = ("N01FUT5", ["h2_H03WTT1"], 0.00049183)

n20C8N3 = ("N20C8N3", ["h2_H05T3WR"], 0.00015611)

n01FUQO = ("N01FUQO", ["h2_H063YJC", "h2_H097TLK", "h2_H074YH2", "h2_H05T3WR",

"h2_H06KOBS"], 0.02648398)

n20C8KV = ("N20C8KV", ["h2_H01FM8E", "h2_H06E9I7"], 0.00004934)

n01FVUA = ("N01FVUA", ["h2_H01FM8E"], 0.00000176)

n01FUOE = ("N01FUOE", ["h2_H01O23Z", "h2_H01P2HN"], 0.00018751)

n01FUSC = ("N01FUSC", ["h2_H05V49E"], 0.00000000)

n01FUJS = ("N01FUJS", ["h2_H01F6M0", "h2_H04PTS0"], 0.00225736)

n01FUB0 = ("N01FUB0", ["h2_H03Q6S0"], 0.00007817)

n20DDHX = ("N20DDHX", ["h2_H03PO3Z", "h2_H03PO1O"], 0.00008176)

n01FUEC = ("N01FUEC", ["h2_H01P2HN"], 0.00010452)

n01FVYL = ("N01FVYL", ["h2_H03WTT1"], 0.00049183)

n20255Y = ("N20255Y", ["h2_H05MF0G"], 0.00000320)

n20DDB4 = ("N20DDB4", ["h2_H05JJHO"], 0.00003570)

n01H2U5 = ("N01H2U5", ["h2_H01F6M0", "h2_H03WUC7", "h2_H01P2HN"], 0.00057702)

n01FVP3 = ("N01FVP3", ["h2_H05V4AW", "h2_H03NG5F"], 0.00009038)

n01FVVP = ("N01FVVP", ["h2_H01F6M0"], 0.00012331)

n01FUKI = ("N01FUKI", ["h2_H01FM8E", "h2_H06E9I7"], 0.00019994)

n01FVTY = ("N01FVTY", ["h2_H01F6M0"], 0.00007046)

n01FVR4 = ("N01FVR4", ["h2_H03WUC7"], 0.00006100)

n01H2R6 = ("N01H2R6", ["h2_H03WUC7"], 0.00005719)

n20DDIT = ("N20DDIT", ["h2_H04PTS0"], 0.00003680)

n01FUV6 = ("N01FUV6", ["h2_H04SBR1", "h2_H01BHXG", "h2_H01O23Z", "h2_H06E9I7"],

0.00551266)

n01H2PP = ("N01H2PP", ["h2_H05V49E"], 0.00000000)

n01FVYX = ("N01FVYX", ["h2_H03P85A"], 0.00779571)

n01FU95 = ("N01FU95", ["h2_H03P85A", "h2_H074YH2"], 0.14710451)

n01FUVF = ("N01FUVF", ["h2_H05V4AW", "h2_H01FM8E"], 0.00008730)

n01FUSM = ("N01FUSM", ["h2_H04SBR1", "h2_H03PO3Z"], 0.00126402)

n01FVQS = ("N01FVQS", ["h2_H01FM8E"], 0.00001851)

n20255X = ("N20255X", ["h2_H05V49E"], 0.00000000)

n01FUU3 = ("N01FUU3", ["h2_H03NGBE", "h2_H01F6M0", "h2_H03NG5F"], 0.00196941)

n01H2GJ = ("N01H2GJ", ["h2_H06Y0NB"], 0.00000296)

n01H2Q3 = ("N01H2Q3", ["h2_H063YJC", "h2_H03P85A", "h2_H097TLK"], 0.00023478)

n01FUOK = ("N01FUOK", ["h2_H01P2HN"], 0.00011846)

n20C8N2 = ("N20C8N2", ["h2_H063YJC"], 0.00320856)

n01FW2N = ("N01FW2N", ["h2_H01FM8E", "h2_H06E9I7"], 0.00001947)

n01H2RE = ("N01H2RE", ["h2_H03WTT1", "h2_H04SBR1"], 0.00063798)

n01FVRO = ("N01FVRO", ["h2_H05V4AW"], 0.00413435)

n01FUGX = ("N01FUGX", ["h2_H05V4AW", "h2_H03NGBE", "h2_H01F6M0", "h2_H03NG5F"],

0.00026048)

n01FVO5 = ("N01FVO5", ["h2_H03P85A"], 0.00012138)

n01FUVH = ("N01FUVH", ["h2_H06Y0NB", "h2_H05V4AW", "h2_H03NGBE"], 0.00359318)

n01FUKF = ("N01FUKF", ["h2_H01O23Z"], 0.00005173)

n01FUHZ = ("N01FUHZ", ["h2_H05TWH4"], 0.00003570)

n01FUOW = ("N01FUOW", ["h2_H063YJC"], 0.00021534)

n01FUPQ = ("N01FUPQ", ["h2_H063YJC", "h2_H03P85A"], 0.00048241)

n01FVO3 = ("N01FVO3", ["h2_H03NG5F"], 0.00000484)

n01FUTE = ("N01FUTE", ["h2_H03WTT1"], 0.00049183)

n01FUTF = ("N01FUTF", ["h2_H03WTT1"], 0.00049183)

n20C8LF = ("N20C8LF", ["h2_H05TWH4", "h2_H063YJC", "h2_H097TLK", "h2_H074YH2",

"h2_H05JJHO", "h2_H06KOBS", "h2_H09DX6M"], 0.08775562)

n01FUV8 = ("N01FUV8", ["h2_H03Q6S0"], 0.00017718)

n01FVOQ = ("N01FVOQ", ["h2_H01BHXG", "h2_H01P2HN"], 0.00030210)

n01FVPV = ("N01FVPV", ["h2_H01BHXG", "h2_H03Q6S0"], 0.00015588)

n01FUOR = ("N01FUOR", ["h2_H05V43Q"], 0.00009544)

n01FVPM = ("N01FVPM", ["h2_H01FM8E"], 0.00000270)

n01FVOF = ("N01FVOF", ["h2_H05V4AW", "h2_H03NGBE", "h2_H01F6M0", "h2_H04PTS0",

"h2_H03WUC7", "h2_H01P2HN"], 0.01504917)

n01FUUL = ("N01FUUL", ["h2_H03PO3Z", "h2_H05V43Q", "h2_H03PO1O"], 0.00168698)

n01FVTS = ("N01FVTS", ["h2_H05V49E"], 0.00000000)

n01FUQ6 = ("N01FUQ6", ["h2_H03WTT1", "h2_H01F6M0", "h2_H03WUC7"], 0.00176322)

146

M. TOMKO

n01FUGV = ("N01FUGV", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000867)

n202578 = ("N202578", ["h2_H01F6M0", "h2_H03NG5F"], 0.00235847)

n20256N = ("N20256N", ["h2_H03P85A", "h2_H03Q6S0"], 0.00036477)

n01FVOH = ("N01FVOH", ["h2_H05V4AW"], 0.00008554)

n01FUBN = ("N01FUBN", ["h2_H05V4AW", "h2_H01FM8E"], 0.00031427)

n01FVRV = ("N01FVRV", ["h2_H03WTT1"], 0.00049183)

n01FVVX = ("N01FVVX", ["h2_H01FM8E"], 0.00001734)

n01FVVL = ("N01FVVL", ["h2_H05MF0G"], 0.00000320)

n01FVUW = ("N01FVUW", ["h2_H01BHXG", "h2_H01P2HN"], 0.00065628)

n01FVPN = ("N01FVPN", ["h2_H03WTT1", "h2_H03WUC7"], 0.00054902)

n01FUTO = ("N01FUTO", ["h2_H05V4AW", "h2_H01F6M0"], 0.00015160)

n20249J = ("N20249J", ["h2_H01O23Z"], 0.00005173)

n01FVLJ = ("N01FVLJ", ["h2_H03WTT1"], 0.00049183)

n01FVM0 = ("N01FVM0", ["h2_H03WTT1", "h2_H03WUC7"], 0.00054902)

n01FVV2 = ("N01FVV2", ["h2_H03WTT1"], 0.00170501)

n01FUT6 = ("N01FUT6", ["h2_H05V49E"], 0.00000000)

n01H2IX = ("N01H2IX", ["h2_H05MF0G"], 0.00000619)

n01FVQH = ("N01FVQH", ["h2_H05RN6G"], 0.00000000)

n207T0T = ("N207T0T", ["h2_H01O23Z", "h2_H06E9I7", "h2_H01P2HN"], 0.00147813)

n20257C = ("N20257C", ["h2_H03WTT1", "h2_H01F6M0"], 0.00111577)

n2025VG = ("N2025VG", ["h2_H01F6M0", "h2_H03WUC7", "h2_H01P2HN"], 0.00286993)

n20C8RN = ("N20C8RN", ["h2_H05MF0G", "h2_H05IWO5", "h2_H06KOBS", "h2_H09DX6M"],

0.00024213)

n20256R = ("N20256R", ["h2_H04SBR1", "h2_H03WUC7"], 0.00020334)

n01FULB = ("N01FULB", ["h2_H01F6M0"], 0.00011450)

n01FVOR = ("N01FVOR", ["h2_H05V49E"], 0.00000000)

n01FULQ = ("N01FULQ", ["h2_H05V49E"], 0.00000000)

n01FVP1 = ("N01FVP1", ["h2_H01F6M0", "h2_H04PTS0"], 0.00009005)

n01FVO6 = ("N01FVO6", ["h2_H05V4AW", "h2_H03NGBE"], 0.00012446)

n01H2PL = ("N01H2PL", ["h2_H05V49E"], 0.00000000)

n01FVTK = ("N01FVTK", ["h2_H05V4AW"], 0.00008554)

n01FVPQ = ("N01FVPQ", ["h2_H01FM8E"], 0.00001488)

n01FUH0 = ("N01FUH0", ["h2_H01FM8E", "h2_H04PTS0"], 0.00026441)

n01FUJX = ("N01FUJX", ["h2_H01BHXG", "h2_H01P2HN"], 0.00015626)

n20256M = ("N20256M", ["h2_H03Q6S0"], 0.00007817)

n01FUJR = ("N01FUJR", ["h2_H01FM8E"], 0.00000469)

n01FVPR = ("N01FVPR", ["h2_H03NGBE"], 0.00008822)

n01FUOF = ("N01FUOF", ["h2_H03PO3Z"], 0.00006884)

n01FVO8 = ("N01FVO8", ["h2_H01FM8E"], 0.00000598)

n01FVYZ = ("N01FVYZ", ["h2_H03WTT1", "h2_H03WUC7"], 0.00054902)

n01H2X0 = ("N01H2X0", ["h2_H06Y0NB", "h2_H03NG5F"], 0.00001507)

n01FUME = ("N01FUME", ["h2_H063YJC", "h2_H03P85A"], 0.00019043)

n01FVQC = ("N01FVQC", ["h2_H03WTT1"], 0.00049183)

n20GR5B = ("N20GR5B", ["h2_H05TWH4", "h2_H063YJC", "h2_H097TLK", "h2_H05RN6G",

"h2_H05JJHO", "h2_H06KOBS"], 0.00920748)

n01FVYM = ("N01FVYM", ["h2_H05V4AW"], 0.00008554)

n01FVU2 = ("N01FVU2", ["h2_H063YJC", "h2_H03P85A"], 0.00054589)

n01FVTQ = ("N01FVTQ", ["h2_H03PNBO"], 0.00015203)

n01FUPS = ("N01FUPS", ["h2_H03PNBO"], 0.00059798)

n01H2FY = ("N01H2FY", ["h2_H063YJC", "h2_H074YH2", "h2_H05V43Q"], 0.00323675)

n01FUEE = ("N01FUEE", ["h2_H063YJC", "h2_H097TLK"], 0.00015202)

n20DDIQ = ("N20DDIQ", ["h2_H097TLK", "h2_H05T3WR"], 0.00071868)

n01H2RB = ("N01H2RB", ["h2_H063YJC", "h2_H074YH2"], 0.00190569)

n20GR5C = ("N20GR5C", ["h2_H063YJC"], 0.00071062)

n01FUOC = ("N01FUOC", ["h2_H06Y0NB", "h2_H03NGBE", "h2_H05V49E"], 0.00029316)

n00E2TA = ("N00E2TA", ["h2_H063YJC", "h2_H097TLK", "h2_H074YH2"], 0.09227029)

n00E643 = ("N00E643", ["h2_H03PNBO", "h2_H05V49E"], 0.00018244)

n00E64D = ("N00E64D", ["h2_H05V49E"], 0.00000000)

n00E64G = ("N00E64G", ["h2_H05V49E"], 0.00000000)

n00E66G = ("N00E66G", ["h2_H06Y0NB"], 0.00000867)

n00E66J = ("N00E66J", ["h2_H05V49E"], 0.00000000)

n00E66L = ("N00E66L", ["h2_H06Y0NB"], 0.00000296)

n01788P = ("N01788P", ["h2_H05V49E"], 0.00000000)

n01789A = ("N01789A", ["h2_H05V49E"], 0.00000000)

n01789B = ("N01789B", ["h2_H05V49E"], 0.00000000)

n01789D = ("N01789D", ["h2_H05V49E"], 0.00000000)

n01789E = ("N01789E", ["h2_H05V49E"], 0.00000000)

n01789F = ("N01789F", ["h2_H05V49E"], 0.00000000)

n01789L = ("N01789L", ["h2_H05V49E"], 0.00000000)

n01789R = ("N01789R", ["h2_H06Y0NB"], 0.00004711)

n01789S = ("N01789S", ["h2_H06Y0NB"], 0.00000729)

n01789T = ("N01789T", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000296)

n01789U = ("N01789U", ["h2_H06Y0NB"], 0.00000296)

n01789V = ("N01789V", ["h2_H05V49E"], 0.00000000)

147


n01789W = ("N01789W", ["h2_H05V49E"], 0.00000000)

n01789Y = ("N01789Y", ["h2_H06Y0NB"], 0.00003430)

n01789Z = ("N01789Z", ["h2_H06Y0NB"], 0.00001577)

n0178A0 = ("N0178A0", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000591)

n0178A1 = ("N0178A1", ["h2_H06Y0NB"], 0.00000315)

n0178A2 = ("N0178A2", ["h2_H06Y0NB"], 0.00001005)

n0178A3 = ("N0178A3", ["h2_H06Y0NB"], 0.00000296)

n0178A4 = ("N0178A4", ["h2_H05V49E"], 0.00000000)

n0178A5 = ("N0178A5", ["h2_H05V49E"], 0.00000000)

n0178A7 = ("N0178A7", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000296)

n0178A8 = ("N0178A8", ["h2_H05V49E"], 0.00000000)

n0178AA = ("N0178AA", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000591)

n01AB6A = ("N01AB6A", ["h2_H05RN6G"], 0.00000000)

n01AB6B = ("N01AB6B", ["h2_H05RN6G"], 0.00000000)

n01AB6C = ("N01AB6C", ["h2_H05RN6G"], 0.00000000)

n01AB6D = ("N01AB6D", ["h2_H05RN6G"], 0.00000000)

n01AB6E = ("N01AB6E", ["h2_H05RN6G"], 0.00000000)

n01AB6G = ("N01AB6G", ["h2_H05RN6G"], 0.00000000)

n01AB6M = ("N01AB6M", ["h2_H05RN6G"], 0.00000000)

n01AB6N = ("N01AB6N", ["h2_H05RN6G"], 0.00000000)

n01AB6O = ("N01AB6O", ["h2_H05RN6G"], 0.00000000)

n01AB6R = ("N01AB6R", ["h2_H05RN6G"], 0.00000000)

n01AB6S = ("N01AB6S", ["h2_H05RN6G"], 0.00000000)

n01ABD8 = ("N01ABD8", ["h2_H05RN6G"], 0.00000000)

n01ABDD = ("N01ABDD", ["h2_H05RN6G"], 0.00000000)

n01ABDF = ("N01ABDF", ["h2_H05RN6G"], 0.00000000)

n01ABDG = ("N01ABDG", ["h2_H05RN6G"], 0.00000000)

n01ABDH = ("N01ABDH", ["h2_H05RN6G"], 0.00000000)

n01ABDI = ("N01ABDI", ["h2_H05RN6G"], 0.00000000)

n01ABDJ = ("N01ABDJ", ["h2_H05RN6G"], 0.00000000)

n01ABDK = ("N01ABDK", ["h2_H05RN6G"], 0.00000000)

n01ABDL = ("N01ABDL", ["h2_H05RN6G"], 0.00000000)

n01ABDM = ("N01ABDM", ["h2_H05RN6G"], 0.00000000)

n01ABDN = ("N01ABDN", ["h2_H05RN6G"], 0.00000000)

n01ABDO = ("N01ABDO", ["h2_H05RN6G"], 0.00000000)

n01ABDP = ("N01ABDP", ["h2_H05RN6G"], 0.00000000)

n01ABDQ = ("N01ABDQ", ["h2_H05RN6G"], 0.00000000)

n01ABDS = ("N01ABDS", ["h2_H05RN6G"], 0.00000000)

n01ABDT = ("N01ABDT", ["h2_H05RN6G"], 0.00000000)

n01ABDW = ("N01ABDW", ["h2_H05RN6G"], 0.00000000)

n01ABDZ = ("N01ABDZ", ["h2_H05RN6G"], 0.00000000)

n01ABE0 = ("N01ABE0", ["h2_H05RN6G"], 0.00000000)

n01ABE1 = ("N01ABE1", ["h2_H05RN6G"], 0.00000000)

n01ABE4 = ("N01ABE4", ["h2_H05RN6G"], 0.00000000)

n01ABE5 = ("N01ABE5", ["h2_H05RN6G"], 0.00000000)

n01ABE7 = ("N01ABE7", ["h2_H05RN6G"], 0.00000000)

n01ABE9 = ("N01ABE9", ["h2_H05RN6G"], 0.00000000)

n01ABED = ("N01ABED", ["h2_H05RN6G"], 0.00000000)

n01ABEO = ("N01ABEO", ["h2_H05RN6G"], 0.00000000)

n01ABES = ("N01ABES", ["h2_H05RN6G"], 0.00000000)

n01ABET = ("N01ABET", ["h2_H05TWH4", "h2_H05RN6G"], 0.00003570)

n01ABEV = ("N01ABEV", ["h2_H05TWH4", "h2_H05RN6G"], 0.00007854)

n01ABEW = ("N01ABEW", ["h2_H05TWH4", "h2_H05RN6G"], 0.00010472)

n01ABEX = ("N01ABEX", ["h2_H05TWH4"], 0.00004046)

n01ABEY = ("N01ABEY", ["h2_H05TWH4"], 0.00003570)

n01ABF0 = ("N01ABF0", ["h2_H05TWH4"], 0.00003570)

n01ABF1 = ("N01ABF1", ["h2_H05TWH4"], 0.00003570)

n01ABF3 = ("N01ABF3", ["h2_H05TWH4"], 0.00010948)

n01ABF4 = ("N01ABF4", ["h2_H05TWH4", "h2_H05JJHO", "h2_H05T3WR"], 0.00007766)

n01ABJ8 = ("N01ABJ8", ["h2_H05RN6G"], 0.00000000)

n01ABJJ = ("N01ABJJ", ["h2_H05RN6G"], 0.00000000)

n01ABJK = ("N01ABJK", ["h2_H05RN6G"], 0.00000000)

n01ABJX = ("N01ABJX", ["h2_H05T3WR"], 0.00002713)

n01ABJY = ("N01ABJY", ["h2_H05T3WR"], 0.00001252)

n01ABJZ = ("N01ABJZ", ["h2_H05T3WR"], 0.00000877)

n01ABK2 = ("N01ABK2", ["h2_H05TWH4", "h2_H05T3WR"], 0.00004476)

n01ABK3 = ("N01ABK3", ["h2_H05T3WR"], 0.00000626)

n01ABK4 = ("N01ABK4", ["h2_H05JJHO", "h2_H05T3WR"], 0.00004196)

n01ABKC = ("N01ABKC", ["h2_H05TWH4"], 0.00006902)

n01ABKF = ("N01ABKF", ["h2_H05TWH4"], 0.00003808)

n01ABKG = ("N01ABKG", ["h2_H05TWH4"], 0.00003570)

n01ABKI = ("N01ABKI", ["h2_H05TWH4"], 0.00010472)

n01ABKN = ("N01ABKN", ["h2_H05TWH4", "h2_H05JJHO"], 0.00007140)

n01ABKO = ("N01ABKO", ["h2_H05TWH4", "h2_H05JJHO"], 0.00007140)

148

M. TOMKO

n01ABL3 = ("N01ABL3", ["h2_H05TWH4", "h2_H05JJHO"], 0.00014280)

n01ABL4 = ("N01ABL4", ["h2_H05TWH4"], 0.00003570)

n01ABL5 = ("N01ABL5", ["h2_H05TWH4"], 0.00014042)

n01ABL6 = ("N01ABL6", ["h2_H05TWH4", "h2_H05JJHO"], 0.00007140)

n01ABL7 = ("N01ABL7", ["h2_H05JJHO", "h2_H05IWO5"], 0.00003570)

n01ABLE = ("N01ABLE", ["h2_H05IWO5"], 0.00000000)

n01ABLF = ("N01ABLF", ["h2_H05IWO5"], 0.00000000)

n01ABOX = ("N01ABOX", ["h2_H063YJC", "h2_H097TLK"], 0.00015202)

n01ABP6 = ("N01ABP6", ["h2_H097TLK"], 0.00009757)

n01ABPC = ("N01ABPC", ["h2_H097TLK"], 0.00004731)

n01ABPD = ("N01ABPD", ["h2_H097TLK"], 0.00004731)

n01ABPM = ("N01ABPM", ["h2_H05JJHO"], 0.00005712)

n01ABPP = ("N01ABPP", ["h2_H05JJHO", "h2_H05T3WR"], 0.00006434)

n01ABPR = ("N01ABPR", ["h2_H05JJHO"], 0.00011186)

n01ABPU = ("N01ABPU", ["h2_H05JJHO"], 0.00003808)

n01ABPW = ("N01ABPW", ["h2_H05JJHO"], 0.00007140)

n01ABPZ = ("N01ABPZ", ["h2_H05JJHO"], 0.00006902)

n01ABQ0 = ("N01ABQ0", ["h2_H05JJHO", "h2_H05T3WR", "h2_H06KOBS"], 0.00007766)

n01ABQ2 = ("N01ABQ2", ["h2_H05JJHO"], 0.00003570)

n01ABQ3 = ("N01ABQ3", ["h2_H05JJHO"], 0.00003570)

n01ABQ6 = ("N01ABQ6", ["h2_H05JJHO"], 0.00010472)

n01ABQ8 = ("N01ABQ8", ["h2_H05JJHO"], 0.00003570)

n01ABQ9 = ("N01ABQ9", ["h2_H05JJHO", "h2_H06KOBS"], 0.00007140)

n01ABQB = ("N01ABQB", ["h2_H05T3WR", "h2_H06KOBS"], 0.00008672)

n01ABQE = ("N01ABQE", ["h2_H06KOBS"], 0.00003570)

n01ABQF = ("N01ABQF", ["h2_H06KOBS"], 0.00004522)

n01ABQG = ("N01ABQG", ["h2_H097TLK", "h2_H06KOBS"], 0.00011207)

n01ABQI = ("N01ABQI", ["h2_H06KOBS"], 0.00003570)

n01ABQJ = ("N01ABQJ", ["h2_H097TLK"], 0.00012714)

n01ABQL = ("N01ABQL", ["h2_H06KOBS"], 0.00003570)

n01ABQP = ("N01ABQP", ["h2_H06KOBS", "h2_H09DX6M"], 0.00011700)

n01ABQQ = ("N01ABQQ", ["h2_H097TLK"], 0.00004731)

n01ABQS = ("N01ABQS", ["h2_H05IWO5", "h2_H09DX6M"], 0.00002647)

n01ABQY = ("N01ABQY", ["h2_H097TLK"], 0.00005913)

n01ABQZ = ("N01ABQZ", ["h2_H09DX6M"], 0.00002482)

n01ABR0 = ("N01ABR0", ["h2_H097TLK"], 0.00008870)

n01ABR1 = ("N01ABR1", ["h2_H097TLK"], 0.00004435)

n01ABV1 = ("N01ABV1", ["h2_H097TLK"], 0.00006505)

n01ABV2 = ("N01ABV2", ["h2_H097TLK"], 0.00004435)

n01ABV3 = ("N01ABV3", ["h2_H097TLK"], 0.00004435)

n01ABV4 = ("N01ABV4", ["h2_H063YJC", "h2_H097TLK"], 0.00034458)

n01ABV5 = ("N01ABV5", ["h2_H097TLK"], 0.00007392)

n01ABVA = ("N01ABVA", ["h2_H097TLK"], 0.00005026)

n01ABVU = ("N01ABVU", ["h2_H097TLK", "h2_H09DX6M"], 0.00011989)

n01ABVV = ("N01ABVV", ["h2_H05MF0G", "h2_H097TLK", "h2_H09DX6M"], 0.00007237)

n01AC10 = ("N01AC10", ["h2_H03PO3Z", "h2_H074YH2"], 0.00053495)

n01ADGP = ("N01ADGP", ["h2_H05V49E"], 0.00000000)

n01ADH6 = ("N01ADH6", ["h2_H06Y0NB"], 0.00000729)

n01ADHA = ("N01ADHA", ["h2_H06Y0NB", "h2_H03NGBE"], 0.00004188)

n01ADHB = ("N01ADHB", ["h2_H06Y0NB"], 0.00000296)

n01ADHC = ("N01ADHC", ["h2_H06Y0NB"], 0.00000296)

n01ADHD = ("N01ADHD", ["h2_H06Y0NB"], 0.00000296)

n01ADHE = ("N01ADHE", ["h2_H06Y0NB"], 0.00000296)

n01ADHH = ("N01ADHH", ["h2_H06Y0NB"], 0.00000296)

n01FUL8 = ("N01FUL8", ["h2_H06KOBS"], 0.00007140)

n01FUSL = ("N01FUSL", ["h2_H03PNBO", "h2_H03NG5F"], 0.01778883)

n01FUTR = ("N01FUTR", ["h2_H04SBR1"], 0.00226050)

n01FUV1 = ("N01FUV1", ["h2_H03PO1O", "h2_H03Q6S0"], 0.00034766)

n01FUVA = ("N01FUVA", ["h2_H03Q6S0"], 0.00027099)

n202566 = ("N202566", ["h2_H03WTT1", "h2_H04SBR1", "h2_H05V43Q"], 0.01227266)

n202568 = ("N202568", ["h2_H03WTT1", "h2_H03PO3Z", "h2_H074YH2", "h2_H05V43Q"],

0.08341847)

n20256D = ("N20256D", ["h2_H074YH2"], 0.04317889)

n20256P = ("N20256P", ["h2_H04SBR1", "h2_H03PO1O"], 0.00122580)

n20256S = ("N20256S", ["h2_H03WTT1", "h2_H03PNBO", "h2_H01F6M0"], 0.05220246)

n20256U = ("N20256U", ["h2_H03PNBO"], 0.00900013)

n207SXP = ("N207SXP", ["h2_H06Y0NB", "h2_H03NGBE"], 0.00008097)

n207SXQ = ("N207SXQ", ["h2_H06Y0NB"], 0.00001597)

n207SXS = ("N207SXS", ["h2_H06Y0NB", "h2_H03PNBO"], 0.00015499)

n207T02 = ("N207T02", ["h2_H05RN6G"], 0.00000000)

n207T5T = ("N207T5T", ["h2_H05JJHO"], 0.00003570)

n207T5V = ("N207T5V", ["h2_H05RN6G", "h2_H05T3WR"], 0.00005844)

n207T5W = ("N207T5W", ["h2_H05JJHO", "h2_H05T3WR"], 0.00005875)

n207T5X = ("N207T5X", ["h2_H05JJHO"], 0.00003570)

149


n20B4C4 = ("N20B4C4", ["h2_H06E9I7"], 0.00001702)

n20C9F0 = ("N20C9F0", ["h2_H04SBR1"], 0.00180255)

n20DDFW = ("N20DDFW", ["h2_H05V4AW"], 0.00009694)

n20DDIH = ("N20DDIH", ["h2_H05V4AW", "h2_H03NGBE", "h2_H03NG5F"], 0.00768900)

n20GR4W = ("N20GR4W", ["h2_H06E9I7"], 0.00000798)

n01FU8L = ("N01FU8L", ["h2_H05RN6G"], 0.00000000)

n01FUCA = ("N01FUCA", ["h2_H05RN6G"], 0.00000000)

n202562 = ("N202562", ["h2_H03P85A", "h2_H03PO3Z", "h2_H074YH2", "h2_H03PO1O"],

0.00000000)

n01HX6G = ("N01HX6G", ["h2_H03WTT1", "h2_H03WUC7", "h2_H01P2HN"], 0.00000000)

n01FULP = ("N01FULP", ["h2_H05MF0G"], 0.00000000)

n01FUT8 = ("N01FUT8", ["h2_H04SBR1", "h2_H03PO3Z", "h2_H05V43Q"], 0.00000000)

n01FVRI = ("N01FVRI", ["h2_H05V43Q"], 0.00000000)

n20GR51 = ("N20GR51", ["h2_H05TWH4"], 0.00000000)

n01FVZA = ("N01FVZA", ["h2_H03PNBO", "h2_H01F6M0"], 0.00000000)

n01FUE6 = ("N01FUE6", ["h2_H05V4AW", "h2_H01FM8E", "h2_H01BHXG", "h2_H04PTS0",

"h2_H01O23Z", "h2_H01P2HN"], 0.00000000)

n01FUFE = ("N01FUFE", ["h2_H05T3WR"], 0.00000000)

n01FVQI = ("N01FVQI", ["h2_H074YH2", "h2_H05V43Q"], 0.00000000)

n01FVR6 = ("N01FVR6", ["h2_H03PO3Z", "h2_H074YH2"], 0.00000000)

n01FVZ8 = ("N01FVZ8", ["h2_H05V4AW", "h2_H03NGBE", "h2_H03PNBO", "h2_H01F6M0"],

0.00000000)

n01FUGU = ("N01FUGU", ["h2_H06Y0NB", "h2_H05V49E"], 0.00000000)

n01FVQ5 = ("N01FVQ5",["h2_H03WTT1"], 0.00000000)

n01FUAZ = ("N01FUAZ", ["h2_H06E9I7"], 0.00000000)

n01FVZ6 = ("N01FVZ6", ["h2_H03Q6S0"], 0.00000000)

n01FUP9 = ("N01FUP9", ["h2_H063YJC", "h2_H03P85A"], 0.00000000)

n01FVPZ = ("N01FVPZ", ["h2_H06Y0NB", "h2_H03NG5F"], 0.00000000)

n01FULE = ("N01FULE", ["h2_H05V4AW", "h2_H01FM8E"], 0.00000000)

n01H2PS = ("N01H2PS", ["h2_H09DX6M", "h2_H09DX6M"], 0.00000000)

150

Appendix C

Program Code

Haskell module Ddesc implementing the model of selection of references for des-tination descriptions.

1 -- created 22/11/2006 , by Martin Tomko

2 -- Test data inputs are separated in data.hs

34 module Destdesc where

5 import Data -- data definitions and input dataset

6 import List -- Haskell system library for list manipulation

78 -- definition of Object classes and associated methods

9 class Elements a where

10 consObject :: a -> Object

11 createWorld :: [a] -> [Object]

1213 instance Elements District where

14 consObject (a,b,c,d,e) = Area a b c d e

15 createWorld [] = []

16 createWorld (s:sx) = if length(s:sx) == 1

17 then [consObject(s)]

18 else [consObject(s)] ++ createWorld(sx)

1920 instance Elements Path where

21 consObject (a,b,c) = Street a b c

22 createWorld [] = []

23 createWorld (s:sx) = if length(s:sx) == 1

24 then [consObject(s)]

25 else [consObject(s)] ++ createWorld(sx)

2627 -- creates World as input data structure , where World = districts/landmarks +

paths (in lists)

28 world :: [Object]

29 world = createWorld(areas) ++ createWorld(paths)

3031 -- transforms route input from district IDs to Objects.

32 route :: [Object]

33 route = createWorld(routeI)

3435 -- common actions for Objects

36 class Objects a where

37 getType :: a -> String

38 fetchLevel :: a -> Level

39 equalLevel :: a -> a -> Bool

40 fetchID :: a -> String

41 fetchName :: a -> String

42 fetchLandmarks :: a -> [ObjectID]

43 fetchSupers :: a -> [Object] -> [Object]

44 fetchExp :: a -> Double

45 fetchMaxLevel :: a -> Level

46 fetchMySuper :: a -> [Object] -> Object

47 fetchNeighbors :: a -> [ObjectID]

4849 instance Objects Object where

151

APPENDIX C. PROGRAM CODE

50 getType (Area _ _ _ _ _) = "Area"

51 getType (Street _ _ _) = "Street"

52 fetchLevel (Area objectID level objectName landmarks neighbors)

53 = level

54 fetchLevel (Street objectID districts expvalue)

55 = error "level cannot be returned for a path"

56 fetchID (Area objectID level objectName landmarks neighbors)

57 = objectID

58 fetchID (Street objectID districts expvalue)

59 = objectID

60 fetchName (Area objectID level objectName landmarks neighbors)

61 = objectName

62 fetchName (Street objectID districts expvalue)

63 = objectID

64 fetchLandmarks (Area objectID level objectName landmarks neighbors)

65 = landmarks

66 fetchLandmarks (Street objectID districts expvalue)

67 = districts

68 fetchExp (Street objectID districts expvalue)

69 = expvalue

70 fetchExp (Area objectID level objectName landmarks neighbors)

71 = error "expvalue cannot be returned for a

district/landmark"

72 fetchNeighbors (Area objectID level objectName landmarks neighbors)

73 = neighbors

74 equalLevel a b = fetchLevel a == fetchLevel b

75 fetchMaxLevel (Area objectID level objectName landmarks neighbors)

76 = fetchLevel (head (ordByLevelDesc [y|y<-world ,( fetchName (Area objectID

level objectName landmarks neighbors)) == (fetchName y)]))

77 fetchMaxLevel (Street objectID districts expvalue)

78 = fetchLevel (fetchCoarsestObjList

79 [y|y<-world , x<-( fetchLandmarks(Street objectID districts expvalue)),

x== fetchID y])

80 fetchSupers (Area objectID level objectName landmarks neighbors) list

81 = [x | x<-list , elem (fetchID (Area objectID level objectName landmarks

neighbors)) (fetchLandmarks x) &&

82 fetchLevel x == (( fetchLevel (Area objectID level objectName landmarks

neighbors))+1)]

83 fetchMySuper (Area objectID level objectName landmarks neighbors) list

84 = if length [x | x<-list , elem (fetchID (Area objectID level objectName

landmarks neighbors))(fetchLandmarks x) &&

85 (fetchLevel x)==(( fetchLevel (Area objectID level objectName landmarks

neighbors))+1) &&

86 (fetchName x)==( fetchName (Area objectID level objectName landmarks

neighbors))]==0

87 then error "no superior with same name"

88 else head [x | x<-list , elem (fetchID (Area objectID level objectName

landmarks neighbors)) (fetchLandmarks x) &&

89 (fetchLevel x)==(( fetchLevel (Area objectID level objectName landmarks

neighbors))+1) &&

90 (fetchName x)==( fetchName (Area objectID level objectName landmarks

neighbors))]

9192 -- main function of the Destination Descriptions model (Tomko , 2007) (wrapper)

93 destDesc :: [Object] -> [Object]

94 destDesc list = nubBy (equalByName)(destDescA (grdDist list) (reverse

(routePaths list)))

9596 -- function combining path references with district/landamrk references by

proper prominence.

97 destDescA :: [Object] -> [Object] -> [Object]

98 destDescA (d:dx)(p:px)

99 | length (d:dx) == 0 = error "no directions"

100 | length (p:px) == 0 = (d:dx)

101 | length [x|x<-( dirConectByProm (subroute d (route))), elem x (selectPaths

route)] /= 0

102 = d:[head [x|x<-( dirConectByProm (subroute d (route))), elem x (selectPaths

route)]]

103 | (fetchLevel (head (ordByLevelAsc (p:px))) <= (fetchLevel(head (ordByLevelAsc

(d:dx)))))

104 = [d]++( destDescA (dx)(p:px))

105 | (equalLevel (head (ordByLevelAsc (p:px))) (head (ordByLevelAsc (d:dx))) ==

True)

106 = (([d]++[p])++( destDescA (dx)(px)))

152

M. TOMKO

107 | otherwise = (d:dx)

108109 -- function for selection of district/landmark references for destination

descriptions

110 -- returns names of districts/landmarks

111 grdDist :: [Object] -> [Object]

112 grdDist route = reverse (makeDistinct xxx)

113 where xxx = (ordByLevelAsc (recDirs route))

114115 -- recursive selection of district/landmark references

116 -- the input is the route

117 recDirs :: [Object] -> [Object]

118 recDirs (s:sx)

119 |getRef s t (s:sx) == t = [t]

120 |otherwise = (getRef s t (s:sx)):( recDirs (subroute (getRef s t (s:sx))

(s:sx)))

121 where t = last (s:sx)

122123 -- function for identification of district/landmark referents

124 -- inputs are start and destination and route

125 -- performs rules 1 and 2

126 getRef :: Object -> Object -> [Object] -> Object

127 getRef s t route

128 | (elem s world && elem t world) == False = error "At least one of the input

objects doesn ’t exist"

129 | (s == t) = t -- error "Start and Destination are the same !!"

130 | otherwise = describe s t route

131132 -- auxiliary function wrapping compareHierarchies

133 describe :: Object -> Object -> [Object] -> Object

134 describe a b route = compareHierarchies (fetchSupersUnique a world)

(fetchSupersUnique b world) route

135136 -- implementation of the topological rules for districts/landmarks

137 compHier :: [Object] -> [Object] -> [Object] -> Object

138 compHier sbranch [] route = error "too coarse input of destination element ,

input finer element"

139 compHier [] tbranch route = error "too coarse input of start element , input

finer element"

140 compHier sbranch tbranch route

141 = if length (t:tx) ==0

142 then error "change to TBT"

143 else

144 if length [x|x<-t, y<-s, isNeighbor x y || x==y ] /= 0

145 then compHier (concat(sx)) (concat(tx)) route

146 else (last t)

147 where (t:tx) = groupBy (equalLevel)(ordByLevelDesc ([t | t<-tbranch , (elem t

sbranch)== False ]))

148 (s:sx) = groupBy (equalLevel)(ordByLevelDesc ([s | s<-sbranch , (elem s

tbranch)== False ]))

149150 -- function retrieving path references

151 -- the output is the list of prominent paths filtered by route context

( compByRouteCTX ), ordered

152 routePaths :: [Object] -> [Object]

153 routePaths a = [fst3 x|x<-( routePathsOrdASC a)]

154155 -- compares paths (in tripples) by their order of occurence along route !! CTX =

context

156 compByRouteCTX :: (Object ,Object ,Object) -> (Object ,Object ,Object) -> Ordering

157 compByRouteCTX (a,b,c) (d,e,f)

158 |( elemIndex c route < elemIndex e route) = LT

159 |( elemIndex b route > elemIndex e route) &&

160 (elemIndex c route == elemIndex f route) = LT

161 |( fetchMaxLevel a < fetchMaxLevel d) = LT

162 |( fetchExp a < fetchExp d) = LT

163 |(( notElem (fetchName b) (grdDistShow route)) ||

164 (notElem (fetchName c) (grdDistShow route))) &&

165 ((elem (fetchName e) (grdDistShow route)) ||

166 (elem (fetchName f) (grdDistShow route)))= LT

167 | otherwise = GT

168169 -- retrieves the list of paths connecting start and destination of a route ,

ordered by prominence

153


170 dirConectByProm :: [Object] -> [Object]

171 dirConectByProm (s:sx)

172 | length (connectedBy s t) >= 1 &&

173 length [x|x<-( connectedBy s t),isEprom x] >= 1 =

174 (ordByPromDesc [x|x<-( connectedBy s t),isEprom x])

175 | length (connectedBy s t) >= 1 &&

176 length [x|x<-( connectedBy s t),isEprom x] == 0 =

177 ([head (ordByPromDesc [x|x<-( connectedBy s t)])])

178 | otherwise = []

179 where t = (last(s:sx))

180181 --AUXILIARY FUNCTIONS

182 -- ordering of path segments by their importance and route context as occuring

along route

183 -- The last in the list is the most prominent

184 routePathsOrdASC :: [Object] -> [(Object ,Object ,Object)]

185 routePathsOrdASC list = ordRTSegments (routePathsT list)

186187 -- ordering of paths segments implementing compByRouteCTX

188 ordRTSegments :: [(Object ,Object ,Object)] -> [(Object ,Object ,Object)]

189 ordRTSegments list = sortBy compByRouteCTX list

190191 -- function filtering paths in context of the route.

192 -- input is route

193 -- output is the list of unique path names and the limit segments of the route

(path , start , destination )

194 routePathsT :: [Object] -> [(Object ,Object ,Object)]

195 routePathsT list = nubBy (equalTrippleBySeg ) (nubBy (equalTrippleByPath )

(filterList (selectPPaths list)))

196197 -- selects prominent paths connecting pairs of elements of the route

198 -- returns the path and the pair

199 selectPPaths :: [Object] -> [(Object ,Object ,Object)]

200 selectPPaths (s:sx)

201 | (length (s:sx)) == 1 = []

202 | otherwise = [(x,s,(head sx))|x<-( connectedBy s (head sx)),isEprom x]

++ selectPPaths sx

203204 -- function filtering path segments so that those closer to destination are

given preference

205 -- (based on filterPaths )

206 filterList :: [(Object ,Object ,Object)] -> [(Object ,Object ,Object)]

207 filterList (s:sx)

208 | length (s:sx) == 1 = [s]

209 | otherwise = (foldl1 (filterPaths) (s:sx)):filterList sx

210211 -- implementation of preference among two triples).

212 filterPaths :: (Object ,Object ,Object) -> (Object ,Object ,Object) ->

(Object ,Object ,Object)

213 filterPaths (x,a,b) (y,c,d)

214 | b==c && x==y = (y,c,d)

215 | otherwise = (x,a,b)

216217 -- final filtering of path triples

218 equalTrippleByPath :: (Object ,Object ,Object) -> (Object ,Object ,Object) -> Bool

219 equalTrippleByPath (a,b,c) (d,e,f)

220 | a==d = True

221 | otherwise = False

222223 -- helping function for filtering of paths (for the nubBy function in

routePaths)

224 equalTrippleBySeg :: (Object ,Object ,Object) -> (Object ,Object ,Object) -> Bool

225 equalTrippleBySeg (a,b,c) (d,e,f)

226 | (b==e && c==f)= True


228229 -- equals object by their name

230 equalByName :: Object -> Object -> Bool

231 equalByName a b

232 | fetchName a== fetchName b = True


234235 -- check all the paths connecting a pair of the elements of the route

236 -- inputs: route , output: list of paths

154

M. TOMKO

237 selectPaths :: [Object] -> [Object]

238 selectPaths (s:sx)

239 | (length (s:sx)) == 1 = []

240 | otherwise = (connectedBy s (head sx))++ selectPaths sx

241242 -- function to check if start and destination of a route are directly connected

by a path

243 dirConect :: [Object] -> Bool

244 dirConect (s:sx)

245 | length (connectedBy objs objt) >= 1 = True


247 where t = (last(s:sx))

248249 --function to retrieve the list of paths connecting two Objects

(Districts/ Landmarks)

250 connectedBy :: Object -> Object -> [Object]

251 connectedBy a b =

252 [x|x<-world , elem (fetchID a)(fetchLandmarks x) &&

253 elem (fetchID b)(fetchLandmarks x) &&

254 isPath x]

255256 -- function to retrieve the subroute of a (including destination )

257 subroute :: Object ->[Object] -> [Object]

258 subroute a list = (subroutex a list)++[( last list)]

259260 subroutex :: Object -> [Object] -> [Object]

261 subroutex a route = [x | x <- route ,(elem a (fetchSupersUniqueDesc x world))]

262263 -- building branches of superordinate objects

264 -- (district , landmark), ordered by level , coarsest first. May have multiple

objects of the same.

265 fetchSupersUniqueDesc :: Object -> [Object] -> [Object]

266 fetchSupersUniqueDesc a list = ordByLevelDesc (fetchSupersUnique a list)

267268 --finds parents of a, including a

269 fetchSupersUnique :: Object -> [Object] -> [Object]

270 fetchSupersUnique a list = makeDistinct (findObjectsX a list)

271272 --finds all superodinate elements of a, including a

273 findObjectsX :: Object -> [Object] -> [Object]

274 findObjectsX a list =

275 if length (fetchSupers a list) == 0

276 then [a]

277 else [a]++ (concat [findObjectsX x list|x<-( fetchSupers a list)])

278279 --test neighborhood of two inputs

280 isNeighbor :: Object -> Object -> Bool

281 isNeighbor a b = elem (fetchID a) (fetchNeighbors b)

282283 -- function returns single object by level from a from list

284 makeDistinctLevel :: [Object] -> [Object]

285 makeDistinctLevel ls = foldl addDistinctLevel [] ls

286287 addDistinctLevel :: [Object] -> Object -> [Object]

288 addDistinctLevel ls x

289 | (fetchLevel x) ‘elem ‘ [fetchLevel y|y<-ls] = ls

290 | otherwise = ls ++ [x]

291292 -- function returns unique values from a list.

293 makeDistinct :: Eq a => [a] -> [a]

294 makeDistinct ls = foldl addDistinct [] ls

295296 addDistinct :: Eq a => [a] -> a -> [a]

297 addDistinct ls x

298 | x ‘elem ‘ ls = ls

299 | otherwise = ls ++ [x]

300301 -- function that orders a list of objects by level. Orders smaller first.

302 ordByLevelAsc :: [Object] -> [Object]

303 ordByLevelAsc areas = sortBy compByLevelAsc areas

304305 -- function that orders a list of objects by level. Orders greater first.

306 ordByLevelDesc :: [Object] -> [Object]

307 ordByLevelDesc areas = sortBy compByLevelDesc areas

155


308309 -- function that compares two objects by level. Orders smaller first.

310 compByLevelAsc :: Object -> Object -> Ordering

311 compByLevelAsc a b

312 | fetchLevel a <= fetchLevel b = LT


314315 -- function that compares two objects by level. Orders greater first.

316 compByLevelDesc :: Object -> Object -> Ordering

317 compByLevelDesc a b

318 | fetchLevel a >= fetchLevel b = LT


320321 -- function that orders paths by descending experiential prominence

322 ordByPromDesc :: [Object] -> [Object]

323 ordByPromDesc pathlist = sortBy compByPromDesc pathlist

324325 -- function that compares paths by experiential prominence. Orders more

prominent first.

326 compByPromDesc :: Object -> Object -> Ordering

327 compByPromDesc a b

328 | fetchExp a >= fetchExp b = LT


330331 -- retrieves first element of a tripple

332 fst3 :: (a,a,a) -> a

333 fst3 (a,b,c) = a

334335 --fetch most granular object in a set

336 fetchCoarsestObjList :: [Object] -> Object

337 fetchCoarsestObjList [] = error "empty list returned by function

fetchCoarsestObjList"

338 fetchCoarsestObjList a = head (ordByLevelDesc a)

339340 -- check of experiential prominence for paths ,

341 defined in module Data

342 isEprom :: Object -> Bool

343 isEprom a = (fetchExp a) >= eimean

344345 -- check of Object type

346 isPath :: Object -> Bool

347 isPath a

348 | getType a =="Area" = False

349 | getType a =="Street" = True

156

Appendix D

Example Test Cases

A suite of example specifications of routes used for testing the model, along withgenerated results.

District and landmark-based test cases:

1. Route from Universitat Hannover to Staatstheater-Oper:

references: [Rathaus, Katasteramt]

route definition: [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H03PO3Z ,

h2_H04SBR1 , h2_H01BHXG , h2_H01P2HN , h2_H04PTS0 , h2_H01FM8E]

referenceIDs: [h5_H06Y0NB ,h2_H04PTS0]



route definition: [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H05V43Q ,

h2_H03PO3Z , h2_H03PO1O , h2_H03Q6S0 , h2_H01O23Z , h2_H04PTS0 ,

h2_H01FM8E]




route definition: [h2_H097TLK , h2_H063YJC , h2_H074YH2 , h2_H03PO3Z ,

h2_H05V43Q , h2_H03WTT1 , h2_H03PNBO , h2_H03NG5F , h2_H01F6M0 ,

h2_H04PTS0 , h2_H01FM8E]


4. Route from Staatstheater-Oper to the Universitat Hannover:

references: [Universitat Hannover]

route definition: [h2_H01FM8E , h2_H04PTS0 , h2_H01F6M0 , h2_H03NG5F ,

h2_H03PNBO , h2_H03WTT1 , h2_H05V43Q , h2_H03PO3Z , h2_H074YH2 ,

h2_H063YJC , h2_H097TLK]

referenceIDs: [h2_H097TLK]

157

APPENDIX D. EXAMPLE TEST CASES

5. Route from Staatstheater-Oper to the Universitat Hannover:

references: [Universitat Hannover]

route definition: [h2_H01FM8E , h2_H04PTS0 , h2_H01F6M0 , h2_H03NG5F ,

h2_H03PNBO , h2_H03WTT1 , h2_H05V43Q , h2_H03PO3Z , h2_H074YH2 ,

h2_H063YJC , h2_H097TLK]

referenceIDs: [h2_H097TLK]

6. Route from Staatstheater-Oper to the Institute of Chemistry:

references: [Universitat Hannover, Institute of Chemistry]

route definition: [h2_H01FM8E , h2_H04PTS0 , h2_H01P2HN , h2_H01BHXG ,

h2_H04SBR1 , h2_H03PO3Z , h2_H074YH2 , h2_H063YJC ,h2_H097TLK ,

h2_H05MF0G]

referenceIDs: [h2_H097TLK , h2_H05MF0G]

Test cases with integrated references to districts, land-

marks and paths:


references: [Rathaus, Katasteramt, Standehausstrasse]

route definition: [h2_H097TLK , h2_H063YJC , h2_H074YH2 ,

h2_H03PO3Z , h2_H04SBR1 , h2_H01BHXG , h2_H01P2HN , h2_H04PTS0 ,

h2_H01FM8E]

referenceIDs: [h5_H06Y0NB , h2_H04PTS0 , N01FUH0]


references: [Rathaus, Katasteramt, Standehausstrasse]


h2_H03PO3Z , h2_H03PO1O , h2_H03Q6S0 , h2_H01O23Z , h2_H04PTS0 ,

h2_H01FM8E]

referenceIDs: [h5_H06Y0NB , h2_H04PTS0 , N01FUH0]

9. Route from Universitat Hannover to Katasteramt:

references: [Rathaus, Katasteramt, Andreasstrasse]


h2_H03PO3Z , h2_H03PO1O , h2_H03Q6S0 , h2_H01O23Z , h2_H04PTS0]

referenceIDs: [h5_H06Y0NB ,h2_H04PTS0 ,N01FVP7]

10. Route from Allianz-Hochhaus to Staatstheater-Oper:

references: [Katasteramt, Karmarschstraße]

route definition: [h2_H05V43Q , h2_H03WTT1 , h2_H01F6M0 ,

h2_H04PTS0 ,h2_H01FM8E]

referenceIDs: [h2_H04PTS0 , N01FUJS]

11. Route from Staatstheater-Oper to Institute for Chemistry:

references: [Universitat Hannover, Im Moore]

158

M. TOMKO


h2_H04SBR1 , h2_H03PO3Z , h2_H074YH2 , h2_H063YJC ,h2_H097TLK ,

h2_H05MF0G]

referenceIDs: [h2_H097TLK , N01FUPO]

12. Route from Staatstheater-Oper to Universitat Hannover:

references: [Universitat Hannover, Bremer Damm]


h2_H04SBR1 , h2_H03PO3Z , h2_H074YH2 , h2_H063YJC ,h2_H097TLK]

referenceIDs: [h2_H097TLK , N00E2TA]

159

APPENDIX D. EXAMPLE TEST CASES

160

References

Agrawala, M., Stolte, C., 2001. Rendering Effective Route Maps: ImprovingUsability Through Generalization. In: SIGGRAPH 2001. ACM, Los Angeles,CA, pp. 241–250.

Alexander, C., 1988. A City is not a Tree. In: Thackara, J. (Ed.), Design AfterModernism: Beyond the Object. Thames and Hudson, London, pp. 67–84.

Allen, G. L., 1997. From knowledge to words to wayfinding: Issues in the pro-duction and comprehension of route directions. In: Hirtle, S., Frank, A. (Eds.),Spatial Information Theory: A Theoretical Basis for GIS. Springer-Verlag,Berlin, pp. 363–372.

Allen, G. L., 1999. Spatial Abilities, Cognitive Maps, and Wayfinding. In:Golledge, R. G. (Ed.), Wayfinding Behavior. Johns Hopkins University PRess,Baltimore, pp. 46–80.

Allen, G. L., 2000. Principles and Practices for Communicating Route Knowledge.Applied Cognitive Psychology 14 (44), 333–359.

Batagelj, V., Mrvar, A., 2006. Pajek—Program for Large Network Analy-sis, v1.12. Available at: http://vlado.fmf.uni-lj.si/pub/networks/pajek/, socialnetwork analysis software.

Bejan, A., 2000. Shape and Structure, from Engineering to Nature. CambridgeUniversity Press, Cambridge, UK.

Bera, R., Claramunt, C., 2003. Topology-Based Proximities in Spatial Systems.Journal of Geographical Systems 5 (4), 331–415.

Bocaletti, S., Latora, V., Moreno, Y., Chavez, M., Hwang, D.-U., 2006. ComplexNetworks: Structure and Dynamics. Physics Reports 424, 175–308.

Caduff, D., Timpf, S., 2005. The Landmark Spider: Representing LandmarkKnowledge for Wayfinding Tasks. In: Barkowsky, T., Freksa, C., Hegarty, M.,Lowe, R. (Eds.), Reasoning with Mental and External Diagrams: Computa-tional Modeling and Spatial Assistance. AAAI Press, Stanford, CA, USA.

Car, A., 1997. Hierarchical Spatial Reasoning: Theoretical Consideration and itsApplication to Modeling Wayfinding. Ph.D. thesis, Department of Geoinfor-mation, Technical University Vienna.

161

REFERENCES

Carrera, F., 1997. Theory of City Form. Campo Santa Maria Formosa, Venice,Italy: A Case Study of the Application of Visual, Dynamic and Scale-InvariantAnalyses for the Description, Interpretation and Evaluation of City Form. Tech.Rep. MIT 11.330, Massachusetts Institute of Technology.

Chown, E., Kaplan, S., Kortenkamp, D., 1995. Prototypes, Location, and Asso-ciative Networks (PLAN): Towards a Unified Theory of Cognitive Mapping.Journal of Cognitive Science 19, 11–51.

Claramunt, C., Winter, S., 2007. Structural Salience of Elements of the City.Environment and Planning B: Planning and Design 34.

Clark, H. H., Carlson, T. B., 1982. Speech Acts and Hearers’ Beliefs. In: Smith,N. V. (Ed.), Mutual Knowledge. Academic Press, London, pp. 1–36.

Clark, H. H., Marshall, C. R., 1981. Definite Reference and Mutual Knowledge.In: Joshi, A., Webber, B., Sag, I. (Eds.), Elements of Discourse Understanding.Cambridge University Press, Cambridge, pp. 10–63.

Clark, H. H., Wilkes-Gibbs, D., 1986. Referring as a Collaborative Process. Cog-nition 22, 1–39.

Conroy Dalton, R., Bafna, S., 2003. The Syntactical Image of the City: A Recip-rocal Definition of Spatial Elements and Spatial Syntaxes. In: Hanson, J. (Ed.),4th International Space Syntax Symposium. University College London, UK,pp. 59.1–59.21.

Couclelis, H., 1996. Verbal Directions for Way-Finding: Space, Cognition andLanguage. In: Portugali, J. (Ed.), The Construction of Cognitive Maps.Kluwer, Dordrecht, pp. 133–153.

Couclelis, H., Golledge, R. G., Gale, N., Tobler, W., 1987. Exploring the Anchor-point Hypothesis of Spatial Cognition. Journal of Environmental Psychology7, 99–122.

Cowan, N., 2001. The Magical Number of 4 in Short-Term Memory: A Recon-sideration of Mental Storage Capacity. Behavioral and Brain Sciences 24 (1),87–185.

Dale, R., 1992. Generating Referring Expressions: Constructing Descriptions ina Domain of Objects and Processes. ACL-MIT Series in Natural LanguageProcessing. MIT Press.

Dale, R., Geldof, S., Prost, J.-P., 2002. Generating More Natural Route De-scriptions. In: Australasian Natural Language Processing Workshop. Canberra,Australia.

Dale, R., Geldof, S., Prost, J.-P., 2003. Coral: Using Natural Language Genera-tion for Navigational Assistance. In: Oudshoorn, M. (Ed.), 26th AustralasianComputer Science Conference (ACSC2003). Adelaide, Australia.

Dale, R., Geldof, S., Prost, J.-P., 2005. Using Natural Language Generation inAutomatic Route Description. Journal of Research and Practice in InformationTechnology 37 (1), 89–105.

162

M. TOMKO REFERENCES

Dale, R., Reiter, E., 1995. Computational Interpretations of ther Gricean Maximsin the Generation of Referring Expressions. Cognitive Science 18, 233–263.

Dalton, N. S. C., 2006. Configuration and Neighborhood: Is Place Measurable?In: Space Syntax and Spatial Cognition Workshop of the Spatial Cognition ’06Conference. Bremen, Germany.

Davies, B., 1995. To Cooperate or Not to Cooperate - it that the Question? In:Edinburgh Linguistics Department Conference ’94.

Denis, M., Pazzaglia, F., Cornoldi, C., Bertolo, L., 1999. Spatial Discourse andNavigation: An Analysis of Route Directions in the City of Venice. AppliedCognitive Psychology 13, 145–174.

Dey, A. K., 2000. Providing Architectural Support for Building Context-AwareApplications. Phd thesis, Georgia Institute of Technology.

Dijkstra, E. W., 1959. A Note on Two Problems in Connection with Graphs.Numerische Mathematik 1, 269–271.

Downs, R. M., Stea, D., 1977. Maps in Minds: Reflections on Cognitive Mapping.New Harper and Row, New York.

Duckham, M., Kulik, L., 2003. “Simplest Paths”: Automated Route Selection forNavigation. In: Kuhn, W., Worboys, M., Timpf, S. (Eds.), Spatial Informa-tion Theory (COSIT 2003). Vol. 2825 of Lecture Notes in Computer Science.Springer-Verlag, Berlin, pp. 169–185.

Edmonds, P. G., 1994. Collaboration on Reference to Objects that are Not Mutu-ally Known. In: 15th International Conference on Computational Linguistics,COLING-94. Kyoto.

Elias, B., 2003. Extracting Landmarks with Data Mining Methods. In: Kuhn,W., Worboys, M., Timpf, S. (Eds.), Spatial Information Theory (COSIT 2003).Vol. 2825 of Lecture Notes in Computer Science. Springer-Verlag, Berlin, pp.398–412.

Elias, B., 2006. Extraktion von Landmarken fuer die Navigation. Phd. thesis,Institut fur Kartographie und Geoinformatik, Universitat Hannover.

Eppell, V., McClurg, B. A., Bunker, J. M., 2001. A Four Level Road Hierarchyfor Network Planning and Management. In: Jaeger, Vicki (Eds.), 20th ARRBConference. Melbourne, Australia.

Figueiredo, L., 2002. Mindwalk. Available at: http://www.mindwalk.com.br,space syntax software.

Figueiredo, L., Amorim, L., 2005. Continuity Lines in the Axial System. In: TheFifth Space Syntax International Symposium. Delft University of Technology,Delft, The Netherlands.

163

REFERENCES

Frank, A., 2000. Spatial Communication with Maps: Defining the Correctness ofMaps Using a Multi-Agent Simulation. In: Freksa, C., Brauer, W., Habel, C.,Wender, K. F. (Eds.), Spatial Cognition II - International Workshop on Mapsand Diagrammatical Representations of the Environment. Vol. 1849 of LectureNotes in Computer Science. Springer-Verlag, Berlin, pp. 80–99.

Frank, A., 2001. Computational Models for Ontology. Tech. rep., Institute ofGeoinformation, TU Vienna, Austria.

Frank, A., 2003. Pragmatic Information Content: How to Measure the Infor-mation in a Route Description. In: Duckham, M., Goodchild, M., Worboys,M. (Eds.), Foundations of Geographic Information Science. Taylor & Francis,London and New York, pp. 47–68.

Frank, A., Bittner, S., Raubal, M., 2001. Spatial and Cognitive Simulation withMulti-Agent Systems. In: Montello, D. R. (Ed.), Spatial Information Theory.Foundations of Geographic Information Science : International Conference,COSIT 2001 Morro Bay, CA, USA, September 19-23, 2001. Proceedings. Vol.2205 of Lecture Notes in Computer Science. Springer-Verlag, Berlin, pp. 124–139.

Freeman, L. C., 1977. A Set of Measures of Centrality Based on Betweenness.Sociometry 40 (1), 35–41.

Freksa, C., Barkowsky, T., 1996. On the Relation Between Spatial Concepts andGeographic Object. In: Burrough, P., Frank, A. (Eds.), Geographic Objectswith Indeterminate Boundaries. Taylor & Francis, London, pp. 109–121.

Freundschuh, S. M., Mark, D. M., Gopal, S., Gould, M. D., Couclelis, H., 1990.Verbal Directions for Wayfinding: Implications for Navigation and GeographicInformation and Analysis Systems. In: Brassel, K., Kishimoto, H. (Eds.), 4thInternational Symposium on Spatial Data Handling. Department of Geography,University of Zurich, Zurich.

Fussell, S. R., Krauss, R. M., 1991. Accuracy and Bias in Estimates of Others’Knowledge. European Journal of Social Psychology 21, 445–454.

Fussell, S. R., Krauss, R. M., 1992. Coordination of Knowledge in Communica-tion: Effects of Speakers’ Assumptions About What Others Know. Journal ofPersonality and Social Psychology 62 (3), 378–391.

Gerrig, R. J., Brennan, S. E., Ohaeri, J. O., 2001. What Characters Know: Pro-jected Knowledge and Projected Co-Presence. Journal of Memory and Lan-guage 44, 81–95.

Gold, C., Angel, P., 2006. Voronoi Hierarchies. In: Raubal, M., Miller, H., Frank,A., Goodchild, M. (Eds.), Geographic Information Science. Vol. 4197 of LectureNotes in Computer Science. Springer-Verlag, Berlin, Germany, pp. 99–111.

Golledge, R. G., Dougherty, V., Bell, S., 1995. Acquiring Spatial Knowledge:Survey versus Route-Based Knowledge in Unfamiliar Environments. Annals ofthe Association of American Geographers 85 (1), 134–158.

164

M. TOMKO REFERENCES

Grice, P., 1957. Meaning. Philosophical Review 66, 377–388.

Grice, P., 1975. Logic and Conversation. In: Cole, P., Morgan, J. L. (Eds.),Speech Acts. Academic Press, New York, pp. 41–58.

Grice, P., 1989. Studies in the Way of Words. Harvard University Press, Cam-bridge, Massachussetts.

Hanneman, R., Riddle, M., 2005. Introduction to Social Network Methods.

Hansen, S., Klippel, A., Richter, K.-F., 2006. Cognitive OpenLS Specification.Tech. Rep. 012-10/2006, University Bremen / University Freiburg.

Haque, S., Kulik, L., Klippel, A., to appear 2007. Algorithms for Reliable Naviga-tion and Wayfinding. In: Barkowsky, T., Freksa, C., Knauff, M., Krieg-Brckner,B., Nebel, B. (Eds.), Spatial Cognition V. Lecture Notes in Computer Science.Springer-Verlag, Berlin.

Hart, P. E., Nilsson, N. J., Raphael, B., 1968. A Formal Basis for the HeuristicDetermination of Minimum Cost Paths. IEEE Transactions on SSC 4, 100–107.

Hasida, K., 1996. Issues in Communication Game. In: 16th International Confer-ence on Computational Linguistics COLING’96. ACM, pp. 531–536.

Hasida, K., Nagao, K., Miyata, T., 1995. A Game-Theoretic Account of Collabo-ration in Communication. In: Lesser, V. (Ed.), ICMAS-95, First InternationalConference on Multi-Agent Systems. AAAI Press; MIT Press, San Francisco,California, pp. 140–147.

Heeman, P. A., Hirst, G., 1995. Collaborating on Referring Expressions. Compu-tational Linguistics 21 (3), 351–382.

Heinzle, F., Anders, K. H., Sester, M., 2006. Pattern Recognition in Road Net-works on the Example of Circular Road Detection. In: Raubal, M., Miller, H.,Frank, A., Goodchild, M. (Eds.), Geographic Information Science. 4th Interna-tional Conference, Giscience 2006, Mnster, Germany, September 20-23, 2006.Proceedings. Vol. 4197. Springer-Verlag, Berlin, pp. 153–167.

Hillier, B., Hanson, J., 1984. The Social Logic of Space. Cambridge UniversityPress, Cambridge, UK.

Hirtle, S., 2003. Neighborhoods and Landmarks. In: Duckham, M., Goodchild,M., Worboys, M. (Eds.), Foundations of Geographic Information Science. Tay-lor & Francis, London and New York, pp. 191–203.

Hirtle, S., Jonides, J., 1985. Evidence of Hierarchies in Cognitive Maps. Memoryand Cognition 13, 208–217.

Hobbs, J. R., 1985. Granularity. In: Joshi, A. K. (Ed.), 9th International JointConference on Artificial Intelligence. Morgan Kaufmann, Los Angeles, CA, pp.432–435.

165

REFERENCES

Ishikawa, T., Montello, D. R., 2006. Spatial Knowledge Acquisition from DirectExperience in the Environment: Individual Differences in the Development ofMetric Knowledge and the Integration of Separately Learned Places. CognitivePsychology 52 (2), 93–129.

Jarvella, R. J., Klein, W., 1982. Speech, Place, and Action. John Wiley & Sons,Chichester, NY.

Jiang, B., Claramunt, C., 2003. A Structural Approach to the Model Generalisa-tion of an Urban Street Network. Geoinformatica 8 (2), 151–171.

Jiang, B., Claramunt, C., 2004. Topological Analysis of Urban Street Networks.Environment and Planning B: Planning and Design 31, 151–162.

Jung, S., Pramanik, S., 2002. An Efficient Path Computation Model for Hierarchi-cally Structured Topographical Road Maps. IEEE Transactions on Knowledgeand Data Engineering 14 (5), 1029–1032.

Kettani, D., Moulin, B., 1999. A Spatial Model Based on the Notions of SpatialConceptual Map and of Objects Influence Areas. In: Freksa, C., Mark, D.(Eds.), Spatial Information Theory. Cognitive and Computational Foundationsof Geographic Information Science: International Conference COSIT’99, Stade,Germany, August 1999. Proceedings. Vol. 1661 of Lecture Notes in ComputerScience. Springer-Verlag, Berlin, pp. 401–416.

Kim, Y. O., Penn, A., 2004. Linking the Spatial Syntax of Cognitive Maps to theSpatial Syntax of the Environment. Environment and Behavior 36 (4), 483–504.

Klein, W., 1979. Wegauskunfte. Zeitschrift fur Literaturwissenschaft und Linguis-tik 33, 9–57.

Klippel, A., 2003a. Wayfinding Choremes. In: Kuhn, W., Worboys, M., Timpf,S. (Eds.), Spatial Information Theory: Foundations of Geographic Informa-tion Science. Vol. 2825 of Lecture Notes in Computer Science. Springer-Verlag,Berlin, pp. 320–334.

Klippel, A., 2003b. Wayfinding Choremes. Conceptualizing Wayfinding andRoute Selection Elements. Phd thesis, University of Bremen.

Klippel, A., Dewey, C., Knauff, M., Richter, K.-F., Montello, D. R., Freksa, C.,Loelinger, E.-A., 2004. Direction Concepts in Wayfinding Assistance Systems.In: Baus, J., Kray, C., Porzel, R. (Eds.), Workshop on Artificial Intelligencein Mobile Systems (AIMS’04). Vol. SFB 378 Memo 84. Universitaet Bremen,Saarbrucken, Germany, pp. 1–8.

Klippel, A., Tappe, H., Habel, C., 2003. Pictorial Representations of Routes:Chunking Route Segments During Comprehension. In: Freksa, C., Brauer,W., Habel, C., Wender, K. F. (Eds.), Spatial Cognition III — Routes andNavigation, Human Memory and Learning, Spatial Representation and SpatialLearning. Vol. 2685 of Lecture Notes in Computer Science. Springer-Verlag,Berlin, pp. 11–33.

166

M. TOMKO REFERENCES

Klippel, A., Winter, S., 2005. Structural Salience of Landmarks for Route Direc-tions. In: Cohn, A., Mark, D. (Eds.), Spatial Information Theory. InternationalConference, COSIT 2005, Ellicottville, NY, USA, September 14-18, 2005. Pro-ceedings. Vol. 3693 of Lecture Notes in Computer Science. Springer-Verlag,Berlin, Heidleberg, pp. 347–362.

Kuhn, W., Raubal, M., 2003. Implementing Semantic Reference Systems. In:Gould, M., Laurini, R., Coulondre, S. (Eds.), AGILE 2003 - 6th AGILE Con-ference on Geographic Information Science. Presses Polytechniques et Univer-sitaires Romandes, Lyon, France.

Kuipers, B., 2001. The Skeleton in the Cognitive Map. A Computational Hypoth-esis. In: 3rd International Space Syntax Symposium. Atlanta.

Kuipers, B., Tecuci, D., Stankiewicz, B. J., 2003. The Skeleton in the CognitiveMap: A Computational and Empirical Exploration. Environment and Behavior35 (1), 81–106.

Lau, I. Y.-M., Chiu, C.-y., 2001. I Know What You Know: Assumptions aboutOthers’ Knowledge and their Effects on Message Construction. Social Cognition19 (6), 587–600.

Levine, W. H., Klin, C. M., 2001. Tracking of Spatial information in Narratives.Memory & Cognition 29 (2), 327–335.

Lovelace, K. L., Hegarty, M., Montello, D. R., 1999. Elements of Good RouteDirections in Familiar and Unfamiliar Environments. In: Freksa, C., Mark, D.(Eds.), Spatial Information Theory. Cognitive and Computational Foundationsof Geographic Information Science: International Conference COSIT’99, Stade,Germany, August 1999. Proceedings. Vol. 1661 of Lecture Notes in ComputerScience. Springer-Verlag, Berlin, pp. 65–82.

Lynch, K., 1960. The Image of the City. The MIT Press, Cambridge, Massa-chusetts, USA.

Maass, W., 1993. A Cognitive Model for the Process of Multimodal, IncrementalRoute Descriptions. In: Frank, A., Campari, I. (Eds.), Spatial informationtheory: A theoretical basis for GIS. Springer-Verlag, Berlin, pp. 1–13.

MacMahon, M., 2005. Understanding and Following Route Instructions ThroughLarge-Scale Space. In: Workshop on Spatial Language and Dialogue.Delmhorst, Germany.

Margolis, E., Laurence, S., 2006. Concepts. In: Zalta, E. N. (Ed.), The StanfordEncyclopedia of Philosophy, spring Edition. The Metaphysics Research Lab,Center for the Study of Language and Information, Stanford University, USA.URL http://plato.stanford.edu/archives/spr2006/entries/concepts/

Mark, D. M., 1986. Automated Route Selection for Navigation. IEEE Aerospaceand Electronic Systems Magazine 1 (9), 2–5.

Marshall, S., 2004. Streets and Patterns. Spon Press.

167

REFERENCES

Michon, P.-E., Denis, M., 2001. When and Why Are Visual Landmarks Usedin Giving Directions? In: Montello, D. R. (Ed.), Spatial Information Theory.Foundations of Geographic Information Science : International Conference,COSIT 2001 Morro Bay, CA, USA, September 19-23, 2001. Proceedings. Vol.2205 of Lecture Notes in Computer Science. Springer-Verlag, Berlin, pp. 292–305.

Miller, G., 1956. The Magical Number Seven, Plus or Minus Two. The Psycho-logical Review 63, 81–97.

Montello, D. R., 1993. Scale and Multiple Psychologies of Space. In: Frank, A.(Ed.), Campari, I. Vol. 716. Springer-Verlag, Berlin, pp. 312–321.

Montello, D. R., 2003. Regions in Geography: Process and Content. In: Duckham,M., Goodchild, M., Worboys, M. (Eds.), Foundations of Geographic Informa-tion Science. Taylor and Francis, London and New York, pp. 173–189.

Montello, D. R., 2005. Navigation. In: Miyake, A., Shah, P. (Eds.), CambridgeHandbook of Visuospatial Thinking. Cambridge University Press, Cambridge,pp. 257–294.

Montello, D. R., Goodchild, M., Gottsegen, J., Fohl, P., 2003. Where’s Down-town?: Behavioral Methods for Determining Referents of Vague SpatialQueries. Spatial Cognition and Computation 3 (2-3), 185–204.

Newman, M. E. J., 2005. Power Laws, Pareto Distributions and Zipf’s Law.Contemporary Physics 46 (5), 323–351.

Nothegger, C., Winter, S., Raubal, M., 2004. Computation of the Salience ofFeatures. Spatial Cognition and Computation 4 (2), 113–136.

Noveck, I., Sperber, D., 2006. The Why and How of Experimental Pragmatics:The Case of ‘Scalar Inferences’. In: Burton-Roberts, N. (Ed.), Advances inPragmatics. Palgrave Macmillan.

Ogden, C. K., Richards, I. A., 2001. Richards, I. A.: Selected Works, 1919-1938.Vol. v.2. Meaning of meaning : a study of the influence of language uponthought and of the science of symbolism. Routledge, London.

Okabe, A., Boots, B., Sugihara, K., Chiu, S. N., 1999. Spatial Tesselations:Concepts and Applications of Voronoi Diagrams, 2nd Edition. Wiley Series inProbability and Statistics. John Wiley and Sons, Ltd., Chichester, UK.

Paraboni, I., van Deemeter, K., 2002. Generating Easy References: the Case ofDocument Deixis. In: Second International Conference on Natural LanguageGeneration (INLG 2002). New York, USA.

Patel, K., Chen, M. Y., Smith, I., Landay, J. A., 2006. Personalizing Routes.In: ACM Symposium on User Interface Software and Technology. ACM Press,Montreux, Switzerland.

168

M. TOMKO REFERENCES

Pattabhiraman, T., Cercone, N., 1990. Selection: Salience, Relevance and theCoupling Between Domain-Level Tasks and Text Planning. In: McKeown,K., Moore, J., Nirenburg, S. (Eds.), 5th International Workshop on NaturalLangauge Generation. Dawson, Pennsylvania, USA.

Penn, A., Hillier, B., Banister, D., Xu, J., 1998. Configurational Modelling ofUrban Transport Networks. Environment and Planning B: Planning and Design25, 59–84.

Peterson, J., Chitil, O., 2005. Haskell.org – The Haskell Home Page. Availableon: http://www.haskell.org, visited on April, 3rd 2007.

Plumert, J. M., Carswell, C., de Vet, K., Ihrig, D., 1995a. The Content andOrganization of Communication about Object Locations. Journal of Memoryand Language 37, 477–498.

Plumert, J. M., Carswell, C., de Vet, K., Ihrig, D., 1995b. The Content andOrganization of Communication about Object Locations. Journal of Memoryand Language 37, 477–498.

Plumert, J. M., Spalding, T. L., Nichols-Whitehead, P., 2001. Preferences for As-cending and Descending Hierarchical Organization in Spatial Communication.Memory and Cognition 29 (2), 274–284.

Porta, S., Crucitti, P., Latora, V., 2006. The Network Analysis of Urban Streets:A Dual Approach. Physica A 369 (2), 853–866.

Raubal, M., Winter, S., 2002. Enriching Wayfinding Instructions With LocalLandmarks. In: Egenhofer, M. J., Mark, D. M. (Eds.), Geographic InformationScience: Second International Conference, GIScience 2002, Boulder, CO, USA,September 25-28, 2002. Proceedings. Vol. 2478 of Lecture Notes in ComputerScience. Springer-Verlag, Berlin, pp. 243–259.

Reiter, E., Dale, R., 1992. A Fast Algotithm for the Generation of ReferringExpressions. In: The 17th ACL (COLING-92). Nantes, France.

Richter, K.-F., 2007a. Context-specific route directions: Generation of cognitivelymotivated wayfinding instructions. Phd dissertation, Department of Mathemat-ics & Informatics, University of Bremen.

Richter, K.-F., 2007b. From Turn-by-Turn Directions to Overview Informationon the Way to Take. In: Gartner, G. (Ed.), Location Based Services andTelecartography. Lecture Notes in Geoinformation and Cartography. Springer-Verlag, Berlin, p. 205214.

Richter, K.-F., Klippel, A., 2005. A Model for Context-Specific Route Directions.In: Freksa, C. (Ed.), Spatial Cognition IV. Vol. LNAI 3343. Springer-Verlag,Berlin Heidelberg, pp. 58–78.

Rogers, S., Fiechter, C.-N., Langley, P., 1999. An Adaptive Interactive Agent forRoute Advice. In: Etzioni, O., Mller, J. P., Bradshaw, J. M. (Eds.), Agents’99: Proceedings of the Third Annual Conference on Autonomous Agents. ACMPress, New York, NY, USA, Seattle, Washington, United States, pp. 198–205.

169

REFERENCES

Rosch, E., 1978. Principles of Categorization. In: Roach, E., Lloyd, B. B. (Eds.),Cognition and Categorization. Lawrence Erlbaum Associates, Hillsdale, NJ,pp. 27–48.

Schelling, T. C., 1960. The Strategy of Conflict. Harvard University Press, Cam-bridge, Massachusetts.

Shannon, C. E., Weaver, W., 1949. The Mathematical Theory of Communication.The University of Illinois Press, Urbana, Illinois.

Shanon, B., 1979. Where Questions. In: 17th Annual Meeting of the Associationfor Computational Linguistics. ACL, University of California at San Diego, LaJolla, California, USA.

Siegel, A. W., White, S. H., 1975. The Development of Spatial Representationsof Large-Scale Environments. Advances in child development and behavior 10,9–55.

Simon, H. A., 1962. The Architecture of Complexity. Proceedings of the AmericanPhilosophical Society 106 (6), 467–482.

Singh, R., 1997. Sketch Planning with GIS. In: UCGIS Annual Assembly.

Smith, B., Varzi, A. C., 2000. Fiat and Bona Fide Boundaries. Philosophy andPhenomenological Research 60 (2), 401–420.

Sorrows, M. E., Hirtle, S., 1999. The Nature of Landmarks for Real and Elec-tronic Spaces. In: Freksa, C., Mark, D. (Eds.), Spatial Information Theory.Cognitive and Computational Foundations of Geographic Information Science:International Conference COSIT’99, Stade, Germany, August 1999. Proceed-ings. Vol. 1661 of Lecture Notes in Computer Science. Springer-Verlag, Berlin,pp. 37–50.

Sperber, D., Wilson, D., 1982. Mutual Knowledge and Relevance in Theories ofComprehension. In: Smith, N. V. (Ed.), Mutual Knowledge. Academic Press,London, pp. 61–85.

Sperber, D., Wilson, D., 1986. Relevance. Basil Blackwell Ltd, Oxford, UK.

Sperber, D., Wilson, D., 2004. Relevance Theory. In: Horn, L., Ward, G. (Eds.),Handbook of Pragmatics. Blackwell, Oxford, pp. 607–632.

Srinivas, S., Hirtle, S., to appear, 2007. Knowledge Based Schematization ofRoute Directions. In: Spatial Cognition V: Reasoning, Action, Interaction:International Conference Spatial Cognition 2006, Bremen, Germany. LectureNotes in Computer Science. Springer-Verlag, Berlin.

Stell, J. G., Worboys, M., 1999. Generalizing Graphs Using Amalgamation andSelection. In: Gueting, R. H., Papadias, D., Lochovsky, F. (Eds.), Advancesin Spatial Databases: 6th International Symposium, SSD’99, Hong Kong,China, July 1999. Proceedings. Vol. 1651 of Lecture Notes in Computer Science.Springer-Verlag, Berlin, pp. 19–32.

170

M. TOMKO REFERENCES

Stevens, A., Coupe, P., 1978. Distortions in Dudged Spatial Relations. CognitivePsychology 10, 422–437.

Taylor, H. A., Tversky, B., 1992. Descriptions and Depictions of Environments.Memory and Cognition 20 (5), 483–496.

Thomson, R. C., Richardson, D. E., 1999. The Good Continuation Principle ofPerceptual Organization Applied to the Generalization of Road Networks. In:ICA 19th International Cartographic Conference. Ottawa, Canada.

Timpf, S., Kuhn, W., 2003. Granularity Transformations in Wayfinding. In:Freksa, C., Habel, C., Brauer, W., Wender, K. (Eds.), Spatial Cognition III– Routes and Navigation, Human Memory and Learning, Spatial Representa-tion and Spatial Learning. Vol. 2685 of Lecture Notes in Computer Science.Springer-Verlag, Berlin, pp. 77–88.

Timpf, S., Volta, G., Pollock, D., Egenhofer, M. J., 1992. A Conceptual Modelof Wayfinding Using Multiple Levels of Abstraction. In: Frank, A., Campari,I., Formentini, U. (Eds.), Theories and Methods of Spatio-Temporal Reason-ing in Geographic Space. International Conference GIS From Space to Ter-ritory: Theories and Methods of Spatio-Temporal Reasoning Pisa, Italy, Sep-tember 2123, 1992 Proceedings. Vol. 639 of Lecture Notes in Computer Science.Springer-Verlag, Berlin, Heidleberg, pp. 348–367.

Tomko, M., August 23, 2004 2004a. Analysis of Wayfinding Scenarios. Internalproject report, CRC-SI Australia.

Tomko, M., 2004b. Case Study - Assessing Spatial Distribution of Web Resourcesfor Navigation Services. In: Claramunt, C., Boujou, A., Kwon, Y. J. (Eds.),4th International Workshop on Web and Wireless Geographical InformationSystems W2GIS 2004. Goyang, Korea.

Tomko, M., Winter, S., 2005. Reconstruction of Scenes from Geo-ReferencedWeb Resources. In: Spatial Science Conference. Spatial Science Institute, Mel-bourne, Australia.

Tomko, M., Winter, S., 2006a. Considerations for Efficient Communication ofRoute Directions. In: Cartwright, W., Yoshida, H., Andrienko, G. (Eds.), IS-PRS Workshop on Spatial Data Communication and Visualization. ISPRS,Vienna, Austria.

Tomko, M., Winter, S., 2006b. Initial Entity Identification for Granular RouteDirections. In: Kainz, W., Riedl, A., Elmes, G. (Eds.), Progress in Spatial DataHandling. 12th International Symposium on Spatial Data Handling. Springer-Verlag, Vienna, Austria, pp. 43–60.

Tomko, M., Winter, S., 2006c. Recursive Construction of Granular Route Direc-tions. Journal of Spatial Science 51 (1), 101–115.

Tomko, M., Winter, S., Claramunt, C., to appear 2007. Experiential Hierarchiesof Streets. Computers, Environment and Urban Systems .

171

REFERENCES

Turner, A., Penn, A., Hillier, B., 2005. An Algorithmic Definition of the AxialMap. Environment and Planning B: Planning and Design 32, 425–444.

Tversky, B., 1993. Cognitive Maps, Cognitive Collages, and Spatial Mental Mod-els. In: Frank, A., Campari, I. (Eds.), Spatial Information Theory A Theoret-ical Basis for GIS. European Conference, COSIT’93 Marciana Marina, ElbaIsland, Italy September 1993, Proceedings. Vol. 716 of Lecture Notes in Com-puter Science. Springer-Verlag, Berlin, pp. 14–24.

Tversky, B., 2003. Points, Planes, Paths, and Portions. In: van der Zee, E.,Slack, J. (Eds.), Representing Direction in Language and Space. No. 01 inExplorations in Language and Space. Oxford University Press, Oxford, pp.132–143.

Wagner, M., Balke, W.-T., Hirschfeld, R., Kellerer, W., 2002. A Roadmap toAdvanced Personalization of Mobile Services. In: Industrial Program of the10th International Conference on Cooperative Information Systems (CoopIS2002). Irvine, CA, USA.

Weissensteiner, E., Winter, S., 2004. Landmarks in the Communication of RouteDirections. In: Egenhofer, M. J., Miller, H., Freksa, C. (Eds.), GeographicInformation Science. Third International Conference, GIScience 2004, Adelphi,MD, USA, October 20-23, 2004. Proceedings. Vol. 3234 of Lecture Notes inComputer Science. Springer-Verlag, Berlin, Heidleberg, pp. 313–326.

Wiener, J. M., Mallot, H. A., 2003. ’Fine-to-Coarse’ Route Planning and Naviga-tion in Regionalized Environments. Spatial Cognition and Computation 3 (4),331–358.

Winter, S., Klippel, A., Tomko, M., May 2, 2005 2005. Deliverable 3.3/2: Exper-iments on Usability. Internal report, CRC-SI Australia.

Winter, S., Raubal, M., Nothegger, C., 2004. Focalizing Measures of Saliencefor Wayfinding. In: Meng, L., Zipf, A., Reichenbacher, T. (Eds.), Map-BasedMobile Services - Theories, Methods and Design Implementations. Springer-Verlag Geosciences, Berlin, pp. 127–142.

Winter, S., Tomko, M., 2004. Shifting the Focus in Mobile Maps. In: Morita,T. (Ed.), Joint Workshop on Ubiquitous, Pervasive and Internet Mapping UP-IMap2004. Tokyo, Japan.

Winter, S., Tomko, M., 2005. Route Directions of Varying Granularity. In: Sem-inar on Spatial Cognition. Schloss Dagstuhl, Germany.

Winter, S., Tomko, M., 2006. Translating the Web Semantics of Georeferences.In: Taniar, D., Wenny Rahayu, J. (Eds.), Web Semantics and Ontology. IdeaGroup Publishing, Hershey, Pennsylvania, USA, pp. 297–333.

Winter, S., Tomko, M., Elias, B., Sester, M., to appear 2007. Landmark Hierar-chies in Context. Environment and Planning B: Planning and Design .

Worboys, M., 2003. Communicating Geographic Information in Context. In:Duckham, M., Goodchild, M., Worboys, M. (Eds.), Foundations of GeographicInformation Science. Taylor & Francis, London and New York, pp. 33–45.

172

M. TOMKO REFERENCES

Zhao, S., 2003. Toward a Taxonomy of Copresence. Presence 12 (5), 445–455.

173

destination descriptions in urban environments phd … · as a true guru, stephan led me through...

Documents