7

Spatial representation: a cognitive view

D M MARK

Approaches to the representation and modelling of geographical phenomena in digital form

have expanded recently with the introduction of new perspectives, notably from cognitive
science. The chapter reviews the roots of this cognitive perspective, and examines the
impacts it is having on GIS. A formal model of topological relations between geographical
objects is introduced, and used as the basis of a simple experiment with human subjects to
show how its concepts are mapped into human cognition. Such experiments show promise
for improved designs for easier-to-use systems.

1 INTRODUCTION concentrate on graphic symbols that can represent

those internal digital representations for
Most computer programs gain their meaning from presentation to and interaction with GIS users.
the relationships between their internal formalisms
and some aspect of the external world. Digital
objects are used to represent entities in the real 2 REALISM AND EXPERIENTIALISM
world, and computational procedures represent real-
world processes. In the science of artificial Many of the characteristics of the geographical
intelligence, a representation has been defined as ‘a world can be determined through objective methods
set of conventions about how to describe a set of such as measurement although human perceptions
things’ (Winston 1984: 21). Winston goes on: ‘A of the geographical world are also established
description makes use of the conventions of a through such cognitive functions as perception,
representation to describe some particular things’. reasoning, and memory. These two worlds, of
The fidelity of these representations of entities and measurement and cognition, are clearly compatible
processes of the world is a key factor in the potential in many aspects. Even if there is no functional
utility and usability of almost any computational connection between them (as a phenomenologist
system. Representation is a key concept in any might assert), experiences provide feedbacks that
application that uses computers. tune internal mental representations to match
This chapter provides a cognitive perspective on reality. This is a restatement of experiential realism,
representations of geographical space, entities, and a philosophical stance outlined and named by
relationships that underpin geographical Lakoff (1987). Experiential realism appears to be a
information systems. Representations of things good base for geographical representation in GIS
geographical that have been developed in the (Couclelis 1988; Frank and Mark 1991; Mark and
cognitive sciences for use in those sciences will not Frank 1996). Also, if mental representations exist,
be a focus here, but will be reviewed for what they and are shaped both by the nature of human brains,
may contain that is relevant to GIS. The review will bodies, and senses and by the nature of the world,
concentrate on how a cognitive science perspective then the same representations should be appropriate
can inform the definition and design of digital both to model spatial cognition and for scientific
objects to represent geography. It will also models of geographical phenomena. Thus the

81
D M Mark

representations discussed in this chapter should not language, and distinctions in spatial situations which
differ systematically from those in other chapters in must be made in order to account for linguistic
this section; rather, different motivations and differences, may reveal fundamentals of spatial
sources of evidence should lead to similar representation that may be useful in GIS as well.
representational systems.

4 ENTITIES, FIELDS, AND PHENOMENA

3 PERCEPTION, BEHAVIOUR, LANGUAGE,
AND COGNITION The first problem in representation is to decide what
is to be represented. What things exist? To put this
‘Cognition’ and ‘cognitive’ are words that refer to more formally, what is the ontology of geographical
conscious thinking, including memory, reasoning, space? This chapter will follow the terminology of
and perception. ‘Perception’ normally has a more the US Spatial Data Transfer Standard (SDTS).
narrow meaning, being used to refer to mental ‘Entities’ are things in the real world; ‘objects’ are
sensations and processes that relate to the senses and things in the digital world. Digital objects and
which occur in the direct presence of sensory stimuli associated attributes and values represent
(this psychological definition of perception differs geographical entities. The distinction between entity
from the common (mis)use of this term in the and object makes explicit the difference between
geographical literature to refer to any mental or things and their representations in a formal system.
cognitive phenomenon, as in ‘environmental
perception’). Some brain functions, and some brain 4.1 Entities versus fields
controls of the body, are outside consciousness –
motor skills are a good example. Such non- Many geographical phenomena are best described
conscious processes are usually excluded from scientifically as fields. Good examples are
‘cognition’. Since human behaviour is a combination topographic elevations, air temperatures, and soil
of perceptual inputs, conscious decisions, and motor moisture content. A 2-dimensional field may be
acts, cognition is just one aspect influencing human defined as any single-valued function of location in a
behaviour. Since people’s reasoning and decisions 2-dimensional space. Fields can be classified
cannot be better than their knowledge of situations, according to the levels of measurement (data types)
cognitive factors often influence behaviour more of the dependent variable. Two of the most
than objective properties of the world such as important types are continuous fields, in which the
distances or directions. dependent variable is measured on an interval or
Language can be considered to be a very ratio scale, and discrete fields, with nominal
specialised form of behaviour, but it also appears to dependent variables. It appears that any
provide a unique window on mental processes. The geographical phenomena can be represented either
relation between language and thought is, however, as a field or as a collection of digital objects.
controversial. Decades ago, amateur linguist However, in many cases one approach is a much
Benjamin Lee Whorf proposed an extreme level of better basis for efficient computation; also, one may
interdependence between language and thought – be a better model of people’s mental representations
that people think in language, and that native of the phenomena. For example, a set of states or
speakers of different languages think (reason) provinces within a country would commonly be
differently (Whorf 1940). In recent times, many represented in GIS as a set of areal objects, or as a
linguists have attacked this conjecture, sometimes set of linear objects that form their boundaries.
known as the Sapir-Whorf Hypothesis, claiming that However, the same phenomenon could also be
it is false, and even ridiculous (cf. Pinker 1994). represented as a discrete 2-dimensional field where
Whether or not language influences thought, it each point in space is mapped into the nominal
seems obvious that thought influences language, and category indicating which state it falls in. Fields can
that studying both the structure and the use of be digitally represented by vector approaches, but
natural language describing spatial situations is a often are represented by raster data structures.
good way to study how people think about such Whether the entity or the field model is more
situations. More specifically, distinctions made by appropriate is particularly interesting for topographic

82
 Spatial representation: a cognitive view

elevations. Topographic data normally are represented knowledge of the cartography and GIS literatures.
in GIS as fields, either through gridded digital elevation For example, in Herskovits’ (1986) study of the
models (DEMs) or as triangular tessellations (TINs: prepositions in English, she uses exactly the same
see Hutchinson and Gallant, Chapter 9). Although classification based on dimensionality; as does
TINs are often placed in the vector GIS class, they still Talmy’s classic work on the relation between
represent a field. However, in discourse, people language and the structure of space (Talmy 1983).
normally communicate about terrain in terms of Evidently, although entities must be 3-dimensional
features such as hills, ridges, valleys, gullies, canyons, to have a real physical existence in a 3-dimensional
etc. Thus, a commonsense GIS, capable of being used world, it is common if not universal for people to
by untrained people, should be able to handle queries conceptualise some geographical entities as points,
about ‘this hill’ or ‘that canyon’. However, this is not others as lines, and still others as areas (regions).
just an issue for easy-to-use GIS: a lot of field data
from the pre-Global Positioning System (GPS) days –
such as the labels on biological specimens in museums 4.3 Entities with uncertain boundaries
– have location specified only by feature names and Whereas some geographical entities have distinct or
descriptions, such as ‘on the east ridge of Sumas crisp boundaries, many lack these and instead are
Mountain’ (see also Seeger, Chapter 30; Shiffer, ‘bounded’ by transition zones (Fisher, Chapter 13).
Chapter 52; and Smith and Rhind, Chapter 47). The frequency of such indistinct boundaries is one
A system to geocode such museum locality data of the most distinctive things about geographical
automatically would have to be able to recognise ridges entities, compared with manipulable (table-top)
and mountains. Technically, either an objects or a fields entities. The frequency of geographical entities with
model could be used internally in a GIS; the other indistinct boundaries has been known for some time,
model could then be presented to users as a ‘view’. yet vector GIS is tuned to represent entities with
crisp boundaries, whereas raster GIS does not
4.2 Dimensionality: points, lines, and areas represent entity boundaries at all. Thus, formal
methods for the representation of geographical
It has become standard practice to classify
entities with uncertain or graded boundaries is an
geographical entities according to their
important new research topic in GIS (cf. Burrough
dimensionality as points, lines, or areas. This
and Frank 1996). Fuzzy set theory represents a
typology is especially prevalent in cartography. For
possible approach to modelling entities with graded
the conventional wisdom of US cartography there is
boundaries, but it has problems (see Fisher, Chapter
arguably no better place to look than the various
13). Fuzzy membership functions can be the
editions of Elements of Cartography by Arthur H
Robinson and his various colleagues at the dependent variables in membership fields, perhaps
University of Wisconsin. In the second edition, represented by rasters, but it is not clear whether a
Robinson (1958: 137) wrote about four kinds of full range of GIS functions can be based on such
geographical quantities (point, line, area, and data or whether results of such implementations
volume) and three kinds of cartographic symbols would be either cognitively acceptable or
(point, line, and area). Robinson discusses scientifically valid. Much work in implementation,
2-dimensional data in a chapter on ‘Mapping testing, and human subjects’ evaluation lies ahead.
quantitative point, line, and area data’, and
separates volume data under the title ‘Mapping 4.4 Entity types
3-dimensional data’. Campbell’s textbook (1984)
follows an almost identical chapter subdivision, with Classification is a fundamental cognitive process,
chapters on ‘Mapping spatial variations: points, and it is widely held that categories lie near the heart
lines, and areas’ and on ‘Mapping spatial variations: of cognition (Lakoff 1987; Rosch 1973, 1978).
surfaces’. It is not surprising that GIS follows the Categories carry a great deal of generic, or default,
same basic classification, using dimensionality as the characteristics of entities and allow people to ‘know’
highest level of subdivision for geographical objects some aspects of novel situations. Geographical
(cf. Fegeas et al 1992). entities are no exception, and classification of
It is interesting that cognitive and linguistic works geographical entities into categories is a well-known
also use such a classification, apparently without process both in everyday thinking and in scientific

83
D M Mark

work. Various subfields of geography have a small number of topological distinctions can lead to
developed elaborate classifications for kinds of a large number of mathematically-distinct topological
landforms, vegetation assemblages, settlements etc. spatial relations. Most work on formalising spatial
Mark (1993) described a cognitive theory of relations, both in GIS and in cognitive science, has
categories and showed how such a theory could relied on researchers’ intuitions as to which relations
form a sound basis for entity types in geographical are worthy of distinction and which are not. Some
data interchange standards such as the US SDTS cognitive principles for motivating formal models of
and other data standards. Using an example of spatial relations are presented in this section (5.1–5.4).
inland water bodies, Mark showed that the
conceptual boundaries between adjacent categories
5.1 Topology and metrics
of water bodies in two closely related European
languages (lake, pond, lagoon in English compared Topology may be defined as ‘those properties of
with lac, étang, lagune in French) did not match up. geometrical figures that are invariant under
The category of water bodies classified as étangs in continuous deformation’ (McDonnell and Kemp
French might be considered to be lakes or ponds or 1995). Many spatial relations between objects are
lagoons in English. In standard English, lakes are topological in nature, including adjacency,
distinguished from ponds mainly by size and containment, and overlap. It has long been known
lagoons are distinguished mainly by their position that topological spatial relationships are learned by
relative to the sea. On the other hand, the distinction humans at a very early age, well under one year
between étang and lac in French seems to be mainly (Piaget and Inhelder 1956). Metric information such
one of water quality with étangs having stagnant as size, shape, distance, or direction can also be very
water. A weakness of Mark’s analysis is that it was important cognitively, but is often used to identify
based on dictionaries and examples, rather than on entity types. Metric aspects of spatial relations often
human subjects’ data. Clearly, further research is refine, rather than define, spatial relations. The
required. The example points to the relation of relative roles of topological and metric properties in
geographical categories to cognition, and to the defining spatial relations in reasoning or language is
potential of cultural differences in geographical complex. Some terms indicate relations which are
entity type definition, which could be an impediment purely topological, and in those cases metric
to cross-cultural geographical information exchange. properties may be irrelevant – ‘within’ and ‘enters’
are probably examples of this. In other cases, metric
properties such as distance or direction, expressed
5 SPATIAL RELATIONS either quantitatively or qualitatively, may determine
the meanings of various terms, for example ‘north
Spatial relations are what distinguish spatial of’ or ‘near’ both normally refine the ‘disjoint’
information from other information. Spatial topological relation, and are ill defined for
relations are often encoded in human natural non-disjoint entities.
language by closed-class linguistic elements,
typically prepositions in western European
5.2 Spatial relations between disjoint objects
languages. ‘Closed-class’ means that there is a
relatively small and fixed set of words (lexicon), and It seems that spatial relations between disjoint entities,
thus a limited number of categories that can be which neither touch nor overlap, are characterised by
distinguished. Spatial relations also can be expressed a system of distinctions that is essentially independent
through verbs that describe trajectories or other of the system used to describe and classify spatial
spatial actions. relations for non-disjoint entities. This section (5.2)
Mathematically, in a metric case, there is an infinite deals with disjoint entities, and the following section
continuum of possible spatial relations between any will deal with the other case.
two entities. Topological characteristics of a situation
are invariant under certain transformations. There is 5.2.1 Distance
evidence that many of the cognitively important Distance may be pure Euclidean distance. In natural
spatial relations are entirely or predominantly language, ‘hedge’ words such as ‘about’ are often
topological. Because of combinatorial principles, even associated with approximate numerical distance.

84
 Spatial representation: a cognitive view

Even more commonly, distance may be given in the library’, which invokes a directional reference
qualitative rather than metric terms, dividing frame centred on and oriented to the library. Systems
distances into perhaps just three categories: ‘at’, to use and understand many spatial terms, especially
‘near’, and ‘far’. There may be gender differences in directional, will have to accommodate multiple
the tendency to use or rely on metric distances, reference frames and switching contexts. Individuals
compared with landmarks. Concepts such as ‘near’ may need to switch among these reference frames
are ill defined, and Robinson and his colleagues during discourse or spatial reasoning. Preferences
(Robinson and Lundberg 1987; Robinson and Wong among multiple available reference frames vary with
1987) worked on calibrating the meaning of ‘near’ situation, scale, and culture, with a few cultures
using fuzzy set theory. typically using cardinal directions even indoors and
for body parts, and with others failing to have
5.2.2 Direction orthogonal geographical coordinate schemes
Direction also may be either qualitative or equivalent to cardinal directions (Pederson 1993).
quantitative. Direction is an orientation specified
relative to some reference frame (see section 5.2.3).
5.3 Spatial relations between non-disjoint objects
Again, in everyday speech, directions seem normally
to be specified qualitatively, typically in either four This section discusses topological spatial relations
or eight directions. For science and navigation, more between entities using a particular formal model
precise metric measures of directions are needed, termed the 9-intersection (Egenhofer and Kuhn, this
normally specified in degrees from some arbitrary volume; Egenhofer and Herring 1994). This model is
direction. Frank (1992) has discussed qualitative based on very simple views of spatial entities: each
reasoning about cardinal directions, and has shown entity is defined to have an interior, a boundary, and
that eight directions are an adequate basis for most an exterior. The exterior fills the ‘universe’, except
spatial reasoning. for the parts of that universe occupied by the entity
Ideally, directional relations are thought of as being itself and its boundary. Uses of the 9-intersection,
between points. Directions are not so straightforward and the discussions of it in this chapter, have been
between spatially-extended entities, since a large range restricted to entities in a 2-dimensional space,
of directions may exist, between any point in one entity although it can also be applied to spaces of fewer or
and any point in the other. Peuquet and Zhan (1987) more dimensions.
discussed this problem and provided heuristic rules for The simple form of the 9-intersection model tests
computing the direction relation between extended each ‘part’ (interior, boundary, exterior) of one
entities. However, there were apparently no human spatial entity against each such part of the other and
subjects tests to evaluate whether the algorithm’s simply records which of the nine possible
conclusions match human intuitions or the usage of intersections (3 × 3) are empty and which are not.
directional terms. The approach advocated in this With two possible ‘states’ for each of nine possible
chapter requires eventual cognitive testing with human intersections, this model could distinguish 29, or 512,
subjects before the heuristics developed by Peuquet and spatial relationships. However, if continuity
Zhan can be considered to be effective and workable constraints are placed on the entities involved, the
representations (see also Peuquet, Chapter 8). number of possible spatial relations is greatly
reduced. The 9-intersection model offers no
5.2.3 Reference frames improvement in explanatory or descriptive power for
Various reference frames are used in discourse and region–region relations, compared with previous
spatial reasoning. Geographically, in many cultures, a models. For a pair of regions, eight spatial relations
reference frame based on cardinal directions seems can be distinguished, as shown in Figure 1. For two
dominant for outdoor (‘geographical’) spaces, simple unbranched line segments, however, the 9-
whereas viewer-centred or object-centred reference intersection model distinguishes 33 different
frames often dominate for bodily or tabletop topological spatial relations. It seems unlikely that
(‘manipulable’) spaces and entities (Mark et al 1987). people distinguish that many relations and we
But even in speech communities that typically use advocate human subjects testing to sort out how
cardinal directions in geographical space, it is these 33 relations are organised in cognitive and
common to describe location as being, say, ‘in front of linguistic systems.

85
D M Mark

grouping task (Mark and Egenhofer 1994b). All of

their experiments so far have been performed with a
geographical context of a road and a park. A park
outline shape was kept constant for all experiments,
and roads were placed in 38 different positions, two
for each 9-intersection relation. Subjects were asked
to put drawings into groups so that the same phrase
or sentence could be used to describe all drawings in
the group. Subjects were native speakers of English,
Mandarin, or German. The experiment produced
two main findings: first, topologically identical
drawings (according to the 9-intersection) were
almost always grouped together by any subjects; but
second, subjects grouped topological relations in a
very wide variety of ways (Mark and Egenhofer
1994b). Beyond the 9-intersection classes, individual
differences dominated the results, and cultural or
Fig 1. The eight distinct spatial relations between two simply linguistic differences, if present, could not be
connected 2-dimensional objects in a 2-dimensional space, detected given the small sample sizes and the within-
according to the 9-intersection model. group variance. Tendencies for relations different in
only one of the nine intersections to be grouped
Line–region relations offer an interesting together were used by Egenhofer and Mark (1995)
intermediate level of complexity and resolution. as part of an evaluation of conceptual
Egenhofer and Herring (1994) showed that 19 neighbourhoods for spatial relations. However,
spatial relations are possible between a line and a different experimental protocols were needed to get
region. These relations have formed the basis for a at more subtle differences in cognitive
program of cognitive evaluation of spatial relations representations of spatial relations.
that Mark and Egenhofer have been conducting over Mark and Egenhofer designed an agreement task
the last several years. This work will be reviewed and a drawing task. In the drawing task, subjects
next as a case study. were presented with blank outlines of a polygon said
to be a park, with a sentence under each drawing.
They were asked to draw a line to represent a road
5.4 Case study: spatial relations between lines that had the spatial relation indicated in the
and regions sentence. This test was applied to 32 English
As noted above, the simple (Boolean) version of the subjects, each of whom was asked to draw 64
9-intersection model distinguishes 19 spatial relations sentences, and 19 Spanish-speaking subjects who
between a simple, unbranched line and a simply drew 43 sentences (Mark and Egenhofer 1995). Such
connected region without holes. Intuition would a protocol tends to identify prototypical
suggest that people would rarely distinguish that configurations. Respondents produced a total of
many spatial relations. However, such intuitions may more than 2800 drawings; 88 per cent of which fell
be wrong, and human subjects testing is essential to into just five spatial relations, roughly equivalent to
determine what distinctions are actually made in ‘inside’, ‘outside’ (disjoint), ‘enters’, ‘crosses’, and
language or spatial reasoning. Tests of users’ ‘goes to’. Sharif (1996) analysed the geometry of the
reactions to spatial relations in queries or reasoning drawings by the English language subjects and found
situations apparently have not been performed. that there were many empirical regularities in the
However, tests using language have been done by spatial relations drawn.
Mark, Egenhofer and colleagues, and this work will In the agreement task, subjects were presented
be summarised here to demonstrate the value of with a road–park map and a sentence (such as ‘the
subjects testing combined with formal models. road crosses the park’) and were asked to evaluate
Mark and Egenhofer first explored the cognitive the degree to which the sentence described the
validity of the 9-intersection relations using a spatial relation between the road and park shown in

86
 Spatial representation: a cognitive view

the map (Mark and Egenhofer 1994a). This task was The results of all this testing has been a general
much more controlled, and produced useful results. It confirmation of the validity of the 9-intersection
has been applied to 12 sentences in English (see Mark model. However, the relations it distinguishes must
et al 1995 for a summary of most of this work) and in most cases be aggregated in order to make
one sentence in Spanish; Abrahamson (1994) also used intuitive spatial relations. Previous work on formal
the same stimuli and experimental protocol for five models of spatial relations has generally assumed,
sentences in Norwegian. A total of more than 550 either explicitly or implicitly, that the relations
subjects have been tested for English sentences, and distinguished form a uniform set of equally distinct
about 90 subjects for the Spanish sentence. When spatial relations. In the new model, the same
responses are averaged for each map–sentence pair, ‘primitive’ spatial relation might be an element in
across all subjects within a language, and rescaled to a two or more cognitive spatial relations. This result
0–1 scale, the values can be interpreted as the requires both a powerful formal model of spatial
‘membership’ of that map in the fuzzy set of all relations, and human subjects testing, using several
configurations described by, for example, ‘the road protocols, to confirm or refine the formal models.
crosses the park’. Correlations between mean
agreements to different sentences, across all maps, can
be interpreted as a measure of the similarity in
6 SUMMARY AND PROSPECTS
meanings of the sentences tested. The mean agreement
values can also be related to other parameters Experiential realism appears to bridge the gap
describing the configurations, in efforts to explain between naive realism and developmental models of
subjects’ responses and perhaps to characterise the cognition. Cognition is related to perception,
meanings of the sentences themselves. language, and behaviour. The exact relation of
The most solid data are for the English sentence
language to cognition is controversial, with some
‘the road crosses the park’ (about 150 subjects) and
believing that language shapes cognition, whereas
the Spanish ‘la carretera cruza el parque’ (about
others see cognition as being independent of
90 subjects). These are plotted in Figure 2, and the
language. Whatever the relation, language is
simple correlations between the means by diagram is
certainly related to culture, and makes an excellent
0.985 (97 per cent of variance in common). Not only
site for the study of certain aspects of cognition.
are the means related to topology in about the same
way, but geometric variations also reduce agreement Both entities and fields exist in cognitive models.
for the English-speaking and Spanish-speaking Entities are typically conceptualised as being
subjects in almost exactly the same way. organised by dimensionality: points, lines, areas,
volumes. Entities are often thought of as having
indistinct boundaries, a fact which is at odds with
1.0 typical GIS representation schemes. Entities are also
categorised, and since many aspects of nature form a
continuum, categories may be relatively arbitrary
0.8 and thus subject to cultural differences. Spatial
relations, on the other hand, seem to be very similar
in disparate cultures and languages. Cognitive
0.6
spatial relations are predominantly topological but
Cruza

metric factors such as distance and direction often

0.4
refine the relations and characterise prototypical
relations. A case study, that of spatial relations
between lines and regions, was used to describe the
0.2 value of research approaches combining formal
models and human subjects testing.
There is a very real sense in which all
0.0 representations are cognitive. Mathematics is, after
0.0 0.2 0.4 0.6 0.8 1.0
all, a formalisation of how at least some people think.
Crosses The cognitive view of spatial relations, however,
Fig 2. Mean agreement values for ‘cruza’ against ‘crosses’. emphasises the importance of human subjects

87
D M Mark

testing, preferably under laboratory controlled (eds) Spatial information theory: a theoretical basis for GIS.
conditions, in defining the nature of the spatial Lecture Notes in Computer Science No. 988. Berlin,
Springer: 553–68
representations that are needed for geographical
information systems and spatial analysis. Mark D M, Egenhofer M J 1994a Calibrating the meanings of
spatial predicates from natural language: line–region
relations. Proceedings, Spatial Data Handling 1994 1: 538–53
References Mark D M, Egenhofer M J 1994b Modelling spatial relations
between lines and regions: combining formal mathematical
Abrahamson F 1994 ‘Uklarhet og noyaktighet i geografiske
models and human subjects testing. Cartography and
informasjonssystemer (fuzziness and accuracy in geographical
Geographic Information Systems 21: 195–212
information systems)’. Unpublished Master’s thesis, Division
of Surveying and Mapping, University of Trondheim Mark D M, Egenhofer M J 1995 Topology of prototypical
(Norway) and Norwegian Institute of Technology spatial relations between lines and regions in English and
Spanish. Proceedings, AutoCarto 12, Charlotte, North
Burrough P A, Frank A U (eds) 1996 Geographic objects with
Carolina, March 1995: 245–54
indeterminate boundaries. London, Taylor and Francis
Mark D M, Frank A U 1996 Experiential and formal models
Campbell J 1984 Introductory cartography. Englewood Cliffs,
of geographic space. Environment and Planning B: Planning
Prentice-Hall
and Design 23: 3–24
Couclelis H 1988 The truth seekers: geographers in search of
Mark D M, Svorou S, Zubin D 1987 Spatial terms and spatial
the human world. In Golledge R, Couclelis H, Gould P (eds)
concepts: geographic, cognitive, and linguistic perspectives.
A ground for common search. Sanata Barbara, The Santa
Proceedings, International Symposium on Geographic
Barbara Geographical Press: 148–155
Information Systems: The Research Agenda, November.
Egenhofer M J, Herring J 1994 Categorising topological Crystal City, Virginia 2: 101–12
spatial relations between point, line, and area objects. In
McDonnell R, Kemp K 1995 International GIS dictionary.
Egenhofer M J, Mark D M, Herring J R (eds) The
Cambridge (UK), GeoInformation International
9-Intersection: formalism and its use for natural-language
spatial predicates. Technical Report 94–1. Santa Barbara, Pederson E 1993 Geographic and manipulable space in two
NCGIA Tamil linguistic systems. In Frank A U, Campari I (eds)
Spatial information theory: a theoretical basis for GIS. Lecture
Egenhofer M J, Mark D M 1995 Modelling conceptual
Notes in Computer Science No. 716. Berlin, Springer: 294–311
neighbourhoods of topological line–region relations.
International Journal of Geographical Information Systems Peuquet D J, Zhan C-X 1987 An algorithm to determine the
9: 555–65 directional relationship between arbitrarily-shaped polygons
in the plane. Pattern Recognition 20: 65–74
Fegeas R G, Cascio J L, Lazar R A 1992 An overview of FIPS
173, the Spatial Data Transfer Standard. Cartography and Piaget J, Inhelder B 1956 The child’s conception of space.
Geographic Information Systems 19: 278–93 London, Routledge
Frank A U, Mark D M 1991 Language issues for GIS. In Pinker S 1994 The language instinct. New York, William
Maguire D J, Goodchild M F, Rhind D W (eds) Geographical Morrow & Co. Inc.
information systems: principles and applications. London, Robinson A H 1958 Elements of cartography, 2nd edition.
Longman/New York, John Wiley & Sons Inc. Vol. 1: 147–63 New York, John Wiley & Sons Inc.
Frank A U 1992a Qualitative spatial reasoning about Robinson V B, Lundberg G 1987 Organisation and knowledge
distances and directions in geographic space. Journal of base considerations for the design of distributed geographic
Visual Languages and Computing 3: 343–71 information systems – lessons from semantic modeling.
Herskovits A 1986 Language and spatial cognition: an Proceedings, International Symposium on Geographic
interdisciplinary study of the prepositions in English. Information Systems: The Research Agenda, November 1987.
Cambridge (UK), Cambridge University Press Crystal City, Virginia 2: 245–55
Lakoff G 1987 Women, fire, and dangerous things: what Robinson V B, Wong R 1987 Acquiring approximate
categories reveal about the mind. Chicago, University of representation of some spatial relations. Proceedings, Eighth
Chicago Press International Symposium on Computer-Assisted Cartography:
604–22
Mark D M 1993 Toward a theoretical framework for
geographic entity types. In Frank A U, Campari I (eds) Rosch E 1973 On the internal structure of perceptual and
Spatial information theory: a theoretical basis for GIS. semantic categories. In Moore T E (ed.) Cognitive
Lecture Notes in Computer Sciences No. 716. Berlin, development and the acquisition of language. New York,
Springer: 270–83 Academic Press
Mark D M, Comas D, Egenhofer M J, Freundschuh Rosch E 1978 Principles of categorisation. In Rosch E, Lloyd
S M, Gould M D, Nunes J 1995 Evaluating and refining B B (eds) Cognition and categorisation. Hillsdale, Erlbaum
computational models of spatial relations through cross- Sharif A R B M 1996 ‘Natural-language spatial relations:
linguistic human-subjects testing. In Frank A U, Kuhn W metric refinements of topological properties’. Unpublished

88
 Spatial representation: a cognitive view

Doctoral dissertation, Department of Spatial Whorf B L 1940 Science and linguistics. Technology Review
Information Science and Engineering, University (MIT): 42. Reprinted in Carroll J B (ed.) 1956 Language,
of Maine thought, and reality: selected writings of Benjamin Lee Whorf.
Talmy L 1983 How language structures space. In Pick H, Cambridge (USA), The MIT Press: 207–19
Acredolo L (eds) Spatial orientation: theory, research, and Winston P H 1984 Artificial intelligence, 2nd edition. Reading
application. New York, Plenum Press (USA), Addison-Wesley

89