Semantic Engineering

Abstract

The chapter shows how minimal assumptions on difficult philosophical questions suffice for an engineering approach to the
semantics of geospatial information. The key idea is to adopt a conceptual view of information system ontologies with a minimal
but firm grounding in reality. The resulting constraint view of ontologies suggests mechanisms for grounding, for dealing
with uncertainty, and for integrating folksonomies. Some implications and research needs beyond engineering practice are discussed.

Full text

Semantic Engineering
Werner Kuhn
Institute for Geoinformatics (ifgi), University of Muenster
Robert-Koch-Str. 26-28, D-48149 Muenster (Germany)
kuhn@uni-muenster.de
Abstract
The chapter shows how minimal assumptions on difficult philosophical
questions suffice for an engineering approach to the semantics of geospa-
tial information. The key idea is to adopt a conceptual view of information
system ontologies with a minimal but firm grounding in reality. The result-
ing constraint view of ontologies suggests mechanisms for grounding, for
dealing with uncertainty, and for integrating folksonomies. Some implica-
tions and research needs beyond engineering practice are discussed.
1 Introduction
Many computer scientists, geographers, geoscientists, cognitive scientists,
philosophers, and knowledge engineers are concerned today with solving
semantic problems posed by data about the environment. Their work has
diverging goals and heated debates often ensue on foundations. For exam-
ple, in Beyond Concepts: Ontology as Reality Representation, Barry Smith
(2004) exposes some confused uses of the term concept in ontology. He
proposes to replace concepts as the subject matter of ontologies by “the
universals and particulars which exist in reality” and goes on to show that
this choice yields a more precise understanding of foundational ontological
relations, such as is-a or part-of. While he demonstrates the value of dis-

64 Werner Kuhn
tinguishing universals and particulars, his arguments do not support aban-
doning the notion of concept, as elusive or abused as it may be. Debates on
universal and particular are older and not easier to settle than those on
concept. For example, the question what it means for particulars (such as
Lake Constance) or universals (such as lake) to “exist in reality” remains
unsettled. Thus, Smith’s critique of concepts is mainly a (justified) expo-
sure of some sloppy language use and modeling.
In defense of concepts, and in an information system context, this chap-
ter advocates a pragmatic stance and an engineering view of semantics. A
vast body of literature on ontology engineering for conceptual modeling
(see, e.g., Guarino and Welti 2002; Guizzardi and Halpin 2008) shows
how productive it can be to avoid throwing out the baby of concepts with
the bathwater of its abuses. I will argue that this is so because
1. information system ontologies are only meant to constrain the use
and interpretation of terms; they do not specify “the meaning” of the-
se terms, much less “the existence” of universals and particulars in
reality;
2. ontological constraint networks are groundable in physical properties
of the environment; for semantics, no other assumptions are needed
about reality.
These two assumptions support a linguistic and an engineering reading of
concepts, make these two views compatible with each other, and anchor
ontologies in reality. They commit to a mind-independent reality, but one
in which no objects, universals or particulars need to be posited, only
stimuli, which humans can detect and build concepts from.
The first assumption recalls Guarino’s characterization of an ontology
as “a set of logical axioms designed to account for the intended meaning of
a vocabulary” (Guarino 1998). However, following the saying that “words
don’t mean, people do” and Putnam’s arguments that meaning is not an
object (Putnam 1975), I consider meaning to be a process. Furthermore, I
treat this process as an engineering artifact. Similar to the processes run-
ning in a chemical plant, meaning processes can then be constrained in
how they run: what people mean when they use a term, and how others in-
terpret the term, can be described and influenced. Dictionaries or feature
attribute catalogues, for example, constrain the uses and interpretations of
words or geodata, respectively.
The second assumption ties ontological constraints to reality. Instead of
a simplistic correspondence between terms and objects in reality, which is
clearly untenable, it suggests a minimal and sufficient grounding of onto-
logical constraint networks in elementary physical properties of the world.
A related paper (Schneider et al. 2009) demonstrates and formalizes this

Semantic Engineering 65
grounding process, drawing heavily on Gibson’s meaningful environment
(Gibson 1986). Here, we will just posit the grounding capability as such
and relate it by analogy to the grounding of geodetic networks.
Based on these two core assumptions, the chapter lays out an engineer-
ing view of semantics. The view has its roots in ontology engineering, but
has a purely semantic purpose. It puts concepts (which are considered to be
always associated with terms) at the center of attention and acknowledges
that their descriptions are necessarily incomplete. Its goal is to enable in-
formation users and providers to constrain the uses and interpretations of
their terms. A semantic engineer designs processes of language use and in-
terpretation.
The chapter first shows that concepts can be treated as symbolic and so-
cial entities subject to constraints (section 2). Then, it explains the result-
ing view of ontologies as constraint networks (section 3), an understanding
of grounding resulting from it (section 4), an integration strategy for folk-
sonomies (section 5), and a mechanism for dealing with uncertainty (sec-
tion 6). It concludes with a discussion of research challenges (section 7).
2 An Engineering View of Concepts
As Smith (2004) states, the lack of convincing definitions of concept and
related terms like conceptualization is partly due to “
the fact that these
terms deal with matters so fundamental to our cognitive architecture
(comparable in this respect to terms like ‘identity’ or ‘object’) that at-
tempts to define them are characteristically marked by the feature of
circularity.
” Replacing concept by universal and particular, however,
does not solve this problem, as the age-old debates on realism, nominal-
ism, and conceptualism show. For solving semantic problems, it may be
more productive to agree on minimal requirements imposed on the notion
of concept. This is what I attempt to do here, limiting foundational claims
to relatively uncontroversial ones, and not attempting a formal definition
of concept. This section states these claims and proposes an interpretation
of the popular semantic triangle capturing them.
2.1 A triadic notion of concepts
For the purposes of semantic engineering, it is necessary and sufficient to
posit a threefold nature of concepts, involving
• terms (symbols, words, expressions),

66 Werner Kuhn
• which evoke and express ideas (thoughts) and
• are used to refer to reality.
For example, in speakers of English, the word lake evokes an idea of a wa-
ter body connected to other water bodies and having properties like a rela-
tively flat surface and a water depth. The word may be used to refer, for
instance, to that large amount of blue, wet substance near Constance, as an
instance of the kind (
lake) or as a named individual (Lake Constance). For
German speakers, the words
See and Bodensee play the same roles, re-
spectively.
Concepts are considered here to be associated with terms in a language,
not detached from them, so that the English word
lake and the German
word
See belong to two different concepts, regardless of whether they are
used to refer to the same parts of reality or not (see also Mark 1993). The
German term Begriff may make this close association between words and
thoughts more explicit than the more abstract English term concept; its
root begreifen (touch) furthermore points to the embodied nature of con-
cepts.
2.2 The Semantic Triangle Revisited
The triadic notion of concepts goes back as far as Aristotle (Sowa 2000)
and is often represented by a semantic (or semiotic, or meaning) triangle
(Ogden and Richards 1923), with one corner for each of the three aspects
(Fig. 1).
Fig. 1. A form of the meaning triangle, adapted from (Ogden and Richards 1923)

Semantic Engineering 67
Note that Ogden and Richards (1923) avoid the term “concept” in their tri-
angle altogether and point to many imprecise uses of the term. For the pur-
pose of semantic engineering, it is sufficient to posit that the corners of the
meaning triangle represent the three aspects of concepts identified above.
Many discussions of the semantic triangle make assumptions about each
corner that are difficult to justify. For example, they label the top corner
“Concept”, claim that the REFERENT corner represents “Objects”, and
define SYMBOLs such as to carry some fixed meaning in and of them-
selves. These and other assumptions about meaning are unnecessarily
strong and probably wrong. The following discussion relaxes some as-
sumptions, makes the remaining ones more precise, and emphasizes the
social embedding of the triangle, which is normally neglected in its discus-
sion.
First, the top corner of the triangle shall, for our purposes, remain a
black box to which semantic engineers have no access, except by assuming
that the use of symbols expresses and evokes some thoughts in some peo-
ple, and that these thoughts are shaped by observations of reality.
Second, the right corner of the triangle is understood here as anything
external to minds that is shared in an information (or language) commu-
nity. It provides physical stimuli, which people can observe and agree on.
Reality, in this context, is what we observe and what we talk about using
symbols. Objects, particulars, and universals do not need to be assumed to
exist in a mind-independent reality; they can be cognitively or socially
constructed. This position does not exclude stronger claims about reality,
but avoids the two-fold circularity of defining universals “as that in reality
to which the general terms used in making scientific assertions corre-
spond” and particulars as “the instances of such universals” in (Smith
2004).
Third, the left corner contains the symbols of the language whose se-
mantics is in question, including symbols denoting relationships (such as
flowing into). For artificial languages, like those of information systems,
one can design these symbols as well as the conditions for their use. This
language design capacity reinforces the idea of semantic engineering.
The edges of the triangle represent relations between the corners, gener-
ated by human activities:
1. People observe reality and form thoughts; for example, the repeated
occurrence of extended horizontal water surfaces may suggest a cate-
gory of lakes.
2. People express thoughts through symbols; for example, one may no-
tice a water surface, while flying over it, and say “we are flying over
a lake”.

68 Werner Kuhn
3. Communication succeeds when others interpret how the symbols re-
fer to reality; for example, a seat neighbor on the plane might respond
“it must be Lake Constance”.
With the edges of the triangle representing many-to-many relations, per-
spectivalism and polysemy are fully admitted, as they should be. Reality
induces all kinds of thoughts, depending on perspectives taken, which are
in turn expressed in multiple ways as symbols, and the symbols are used to
refer to reality in many ways, even within a single language community.
The triadic notion of concepts avoids their reduction to purely linguistic
or purely mental entities. The symbolic and mental sides are necessary and
inseparable components of concepts and the reference to reality grounds
them. The concepts constructed by a language community are not arbi-
trary, and they are not just entities created by modelers. Rather, they are
what Smith calls “
tools (analogous to telescopes or microscopes) which
we can use in order to gain cognitive access to corresponding entities in
reality” – except that some of the corresponding entities are mentally
and socially constructed, while others are directly observable.
Finally, the proposed view of concepts also acknowledges their social
aspects. In information systems, as well as in communication in general,
terms can only be used to refer to something if there is a language commu-
nity establishing and sustaining this use. The semantic triangle does not
make this social aspect explicit, which is one of its weaknesses. Implicitly,
however, all its corners and edges require concepts to be situated in a
community sharing a language (or parts of it).
3 Ontologies as Networks of Constraints
Ontologies, in our semantic engineering view, constrain the use and inter-
pretation of terms in an information community. For example, a hydrology
ontology constrains how terms like
lake or waterbody should be used and
interpreted. The non-logical symbols of an ontology stand for concepts and
relations and its logical sentences constrain these. For instance, the con-
stants lake and waterbody in the sentence
lake is-a waterbody
stand for the concepts
lake and waterbody, respectively. By committing to
the ontology, a hydrological information community constrains these two
concepts through the is-a relation. The consequence is that anything
stated about water bodies applies also to lakes. For example, the ontology
could state that

Semantic Engineering 69
waterbody has-a waterdepth
and thereby also constrain all lakes to have a water depth quality. By add-
ing more and more sentences, such as
river is-a waterbody
a network of constraints is incrementally being built up, narrowing the
possible interpretations and uses of terms.
Some symbols may be introduced in the ontology for completeness or
convenience, without necessarily expressing a domain notion. For exam-
ple, in a sentence
every river flows_into a waterbody
the relation flows_into may express a notion of flowing into used in the
hydrology community, but it may also be an auxiliary concept used in the
ontology only.
While the notion of concept is typically reserved for universals in the
literature, ontologies can also constrain terms for particulars, in sentences
like
LakeConstance instance-of lake
or
Rhine flows_into LakeConstance.
This generalization allows for reasoning on individuals in the ontology, not
just in a database or GIS, where this kind of reasoning is typically (and of-
ten more efficiently) performed. Gazetteers are a good example for the
need of a combined reasoning on universals and particulars (Janowicz and
Keßler 2008). Also, ontological specifications of geographic kinds, like
lake or mountain, may refer to the (individual) surface of the Earth, of
which all their instances are parts.
The semiotic function of ontologies themselves (representing concepts
in logical languages) does not require the second meaning triangle that is
sometimes proposed (Sowa 2000). The symbols of the ontology can be
taken to be the symbols of the object language (for example, of hydrology
terms), or syntactic variants of them, expressing the same thoughts and re-
ferring to the same reality.
4 Grounding Constraint Networks
Treating ontologies as networks of constraints can give us an understand-
ing of what it means to ground them. The nodes and edges of a network of

70 Werner Kuhn
concept specifications can be further constrained by observations. For ex-
ample, the node lake in the ontology can be tied to polygons representing
lakes in a GIS database, as proposed in (Bennett et al. 2008), or the node
waterdepth can be tied to an observation procedure, as in (Schneider et al.
2009). Such observational information is then propagated through the net-
work and further restricts possible interpretations of all connected terms. If
it is supplied in symbols that are grounded in physical reality, this ground-
ing propagates through the network.
Grounding is a process of adding information on the variables of a con-
ceptual network through observations anchored in physical stimuli. This
idea concurs with Quine’s notion of observation sentences (Quine 1960).
Anchoring typically occurs in measurement units, other reproducible con-
ventions about measurement (such as agreements on zero values), and fun-
damental observable properties of the environment (like the fact that two
different media are separated by a surface). While a complete theory of on-
tology grounding remains to be worked out, I will explain the main idea
here using a geodetic analogy. A worked out example and the relation to
environmental psychology are presented in (Schneider et al. 2009).
Geodesists are familiar with the idea of grounding a constraint network:
using triangulation networks, they compute coordinates from observations
of distances and directions. The distance and directions are expressed as
constraints, which are parameterized in the coordinates and thereby map
the coordinate space to an observation space (Vaníček et al. 1982). The
networks are grounded through measurement units and externally supplied
coordinate values, which are both anchored in the physics of the earth.
The grounding of triangulation networks is called a geodetic datum. It
ties the social constructions of coordinate systems (in particular, their
equator and zero meridian) to the body of the earth. Broadly speaking, the
earth’s shape determines the ellipsoid on which the coordinates are de-
fined, the mass center anchors it in space, and the rotation axis orients it.
The essence of this grounding scheme is to achieve reproducible interpre-
tations of coordinates: one can take any coordinates and reconstruct the
corresponding real-world location, at least in principle.
A geodetic datum determines the interpretations of the coordinate con-
cepts used to describe location, i.e. of coordinates as such, not just of par-
ticular coordinate values. It explains the notions of latitude and longitude
operationally, by giving a recipe of how they are measured. Seen as net-
works constraining concepts, triangulation networks constrain the interpre-
tation of coordinates, distances and directions. Practically, these pose no
semantic problems, since methods exist to compute the necessary interpre-
tations. For coordinates, these methods are the coordinate reference sys-
tems (ISO 2002) commonly used in GIS and other geospatial information

Semantic Engineering 71
technology. For distance and direction measurements, they are the SI sys-
tem of measurement units (SI). The former, of course, are themselves an-
chored in the latter.
Conceptually, this interpretation procedure for coordinates supplies an
analogy for interpreting terms like
lake or waterdepth. Both kinds of inter-
pretation processes, geodetic and general, first map the symbols to obser-
vations and then ground this mapping in physical reality. Such an analogy
is at the heart of the notions of a semantic reference system and semantic
datum introduced in (Kuhn 2003; Kuhn and Raubal 2003). Here, we have
extended it to a constraint view of ontologies, to clarify the notion of on-
tology grounding.
It should be noted that grounding can never be absolute. A geodetic da-
tum rests on geophysical models (e.g., for the mass distribution of the
earth) and on astronomical frames of reference (star positions). Strictly
speaking, these assumptions harm the reproducibility of coordinate posi-
tions. More generally, a semantic datum can only shift the need for inter-
pretation to a reference frame at the next level. Practically, this shift should
(by design) solve most semantic problems. Philosophically, however, the
caveat may be useful to consider by ontologists making stronger assump-
tions about reality.
5 Integrating Folksonomies with Ontologies
A further gain of a constraint view of ontologies is that it connects ontolo-
gies to folksonomies. Folksonomies are non-hierarchical lists of keywords
(tags) linked to information resources. For example, a web site describing
a bicycle tour around Lake Constance
1
has been tagged with the terms cy-
cle
, tour, austria, bregenz, lake, constance by a user of the social book-
marking site delicious.com
2
.
Folksonomies are not related to taxonomies, despite their name, but
provide data about the terms people associate with contents. They do not
contain logical axioms, but tuples linking terms to resource identifiers (and
to the tagging users). These tuples constrain interpretations of the terms
used as tags, by showing their use and its evolution over time. They con-
strain the interpretations bottom-up, complementing the prescriptive top-
down constraints of ontologies. For example, delicious.com reveals what
contents are tagged by terms like
lake and/or river.
1
http://www.bicyclegermany.com/lake_constance.htm
2
http://delicious.com/url/a8dccabc65ed02711e150a743f226fff?show=all

72 Werner Kuhn
Folksonomies are easy to generate and use, do not burden users or pro-
ducers with difficult modeling tasks, and clearly have something to tell us
about the semantics of their terms. They are, in fact, an increasingly popu-
lar form of empirical semantic data. Other such forms come from data
mining and similar knowledge extraction methods. All these inductive ap-
proaches to semantics play a key role in the automated learning and main-
tenance of terminological constraint systems.
6 Accommodating Uncertainty
Constraint networks provide great flexibility in information handling, by
admitting any number of constraints on any of their variables. As a conse-
quence, they have to provide methods to accommodate uncertainty, since
the stated constraints may over- or underdetermine an exact solution. In
the case of triangulation networks, as well as for many other cases, one
considers all variables and observables to be stochastic, i.e., having a prob-
ability density distribution (Vaníček 1982). Relative weights on the con-
straints can then be derived from knowledge or assumptions about this dis-
tribution, i.e., about the precision of the constraints.
Zadeh’s Generalized Constraint Language (Zadeh 2008) provides the
formal framework to extend this idea to general cases of “computing with
words”. Perception-based statements, i.e., observations, can be precisiated
by any suitable means (for example, by probability distributions or fuzzy
set membership curves), and their impact on a solution for the network
variables can be computed.
This methodology of Zadeh bridges the traditional two-valued logic on-
tologies to the constraint-based view suggested here, where semantic in-
formation is by default considered to be uncertain. The main difference to
geodetic or other geometrically well-defined cases is that conceptual net-
works have no clear-cut degrees of freedom. Grounding can therefore not
easily be determined to be sufficient, but the geodetic ideas that
• grounding spreads through the network;
• arbitrary observational information can be added;
• assumptions on the relative precision of this information serve to weigh
its impact;
remain valid in the “computing with words” scenario of semantic engi-
neering.

Semantic Engineering 73
7 Conclusions
Semantic engineering constrains interpretations of terminologies. It im-
proves mechanisms for information sharing by using semantic web and so-
cial web technology to formulate and evaluate constraints on interpreta-
tions. It makes only minimal assumptions about difficult philosophical
issues (reference, realism vs. nominalism, cognitive processes), in order to
allow for pragmatic solutions to semantic problems.
The proposed engineering view of semantics avoids the pitfalls of treat-
ing ideas decoupled from language (and thereby treading on thin ice re-
garding testability of its hypotheses) or treating terminology decoupled
from its use (and thereby limiting semantics to linguistic relations). In-
stead, it rests on a notion of concepts that necessarily involves expressions
in a language and ties them to reality. It specifies concepts in constraint
networks and grounds ontological constraints in observations of reality.
Thereby, it admits conceptual theories, but avoids engaging in psychologi-
cal speculation about what ideas people may have about the world. The
matter of study (and of engineering design) is how people apply terms to
refer to something in the world that is either commonly observable in an
information community or traceable to something that is.
What these observable aspects exactly are is a question discussed else-
where (Schneider et al. 2009). It constitutes one of the core research ques-
tions raised by semantic engineering. Frank has proposed ontological tiers
to capture references to reality at multiple levels of abstraction (Frank
2001). Here, I refrain from assuming anything about such levels (e.g. about
their ordering or about the role of objects) and only posit that observation
sentences (in the sense of Quine) exist, so that primitive symbols can be
interpreted through ostension. While there may be philosophical quibbles
against this position as a general requirement, it appears to rest on solid
ground in the context of geospatial information, which is per definition
rooted in observations of the environment.
Ontology research has made limited use of the idea that ontologies are
networks of constraints on concepts. Yet, concept networks have been a
central idea in dealing with semantics for a long time, both in linguistics
(Langacker 1987) and in computing (Woods 1985). Networks of con-
straints are a standard device in many areas of engineering and computing.
The chapter has shown that ontologies seen as constraint networks supply
mechanisms for grounding, for accommodating uncertainty, and for inte-
grating folksonomies.
Mechanisms for the formal treatment of ontological constraint networks,
including their grounding and uncertainty, remain to be refined and im-

74 Werner Kuhn
plemented. It appears that model theory is a sufficient formal basis, if ob-
servable aspects of reality are admitted as models, as proposed, for exam-
ple, in (Hayes 1985). These models are then algebraic, consisting of ob-
servable qualities and their changes, and transcend the naïve set-based
model theory of formal semantics.
Picking out one symbol and considering only one sense of it turns the
relations at the edges of the meaning triangle into functions. This allows
for a categorical formalization of the triangle, where the refer function is
treated as a composition of observe and express functions. Thereby, se-
mantic theories may get connected to theories of change and action (Kuhn
2005), explaining semantics through observable effects of processes. For
example, this would make it possible to explain why the seat neighbor in
the flight over Lake Constance might leave his seat after the above dia-
logue to stretch his legs before an expected landing in Zurich.
The combination of linguistic, mental, empirical, and social aspects of
concepts advocated here allows for constraining how information produc-
ers and consumers interpret terms. It permits agreements on such interpre-
tations in the form of ontologies and it can deal with their evolution over
time. This pragmatic position has nothing to do with “cultural relativism”.
It rests on the basic scientific paradigm of knowledge derived from obser-
vations. It is compatible with, but does not require, a stronger form of real-
ist semantics, but avoids some pitfalls of both, realist and cognitive seman-
tics. For example, it has no need to invoke truth independently of meaning,
or to decide which entities have correspondents in reality and which not,
nor does it have to assume unverifiable cognitive mechanisms. The only
cognitive claim is that humans interpret the terms they use, and that this in-
terpretation is ultimately based on ostension. An appropriate philosophical
basis for such a view is radical constructivism (Glasersfeld 2002), which
treats human conceptualizations and knowledge as constructions, con-
strained by observations and interactions with other individuals and with
the environment.
Smith’s program of ontology as science (Smith 2008) is compatible
with, but cannot replace an engineering approach; at least not in the con-
text of geospatial information, which exhibits multiple and often conflict-
ing conceptualizations of reality. A single reference ontology (which
Smith pursues for biomedical information) is unlikely to emerge any time
soon for geospatial domains.
Meanwhile, putting application terminology on a solid basis in the form
of foundational ontologies (such as DOLCE, BFO, CIDOC-CRM, SUMO)
helps making sensible general distinctions. For example, universals and
particulars, endurants and perdurants, or different types of qualities are
usefully distinguished, though the distinction ultimately rests on human

Semantic Engineering 75
conceptualizations. DOLCE’s distinctions suggest a notion of primitive
qualities, which support grounding in observations. This kind of anchoring
of ontologies is likely to support ontology mappings at least as effectively
as the abstract scientific notions from a reference ontology, which are al-
most guaranteed to be interpreted differently in multiple applications.
Acknowledgments
Support for this work was provided in part by the European Commission
(IST project SWING, No. FP6-26514). Many discussions in MUSIL
(http://musil.uni-muenster.de) and with other colleagues (Andrew Frank,
David Mark, Boyan Brodaric and others) as well as comments from two
anonymous reviewers have shaped the ideas expressed and helped me ar-
ticulate them. I am grateful to Lotfi Zadeh for asking how my approach
handles uncertainty, to Andrew Frank for asking where objects come from,
and to Mike Worboys for asking why we need to talk about anything be-
yond terms. The chapter gives some preliminary answers.
References
Bennett B, Mallenby D, Third A (2008) An Ontology for Grounding Vague
Geographic Terms. In: Eschenbach C, Grüninger M (eds) Formal Ontology in
Information Systems (FOIS 2008), IOS Press, pp 280–293
Frank AU (2001) Tiers of ontology and consistency constraints in geographical
information systems. International Journal of Geographical Information
Science (IJGIS) 15(7): 667–678
Gibson JJ (1986) The Ecological Approach to Visual Perception, LEA Publishers,
Hillsdale, NY
Glasersfeld Ev (2002) Radikaler Konstruktivismus: Ideen, Ergebnisse, Probleme
Guarino N (1998) Formal Ontology and Information Systems. In: Guarino N (ed)
Formal Ontology in Information Systems (FOIS’98). IOS Press, Amsterdam,
Trento, Italy, pp 3–15
Guarino N, Welty C (2002) Evaluating Ontological Decisions with
ONTOCLEAN. Communications of the ACM 45: 61–65
Guizzardi G, Halpin T (eds) (2008) Special Issue: Ontological Foundations for
Conceptual Modeling. Applied Ontology 3(1-2)
Hayes PJ (1985) The Second Naive Physics Manifesto. In: Moore Ha (ed) Formal
Theories of the Common-sense World. Ablex, Norwood, NJ, pp 1–36
ISO (2002) ISO 19111 - Spatial referencing by geographic coordinates, ISO TC
211

76 Werner Kuhn
Janowicz K, Keßler C (2008) The Role of Ontology in Improving Gazetteer
Interaction. International Journal of Geographical Information Science
(IJGIS) 22(10): 1129–1157
Kuhn W (2003) Semantic Reference Systems. International Journal of Geographic
Information Science (IJGIS, Guest Editorial) 17: 405–409
Kuhn W (2005) Geospatial Semantics: Why, of What, and How? Journal on Data
Semantics: 1–24
Kuhn W, Raubal M (2003) Implementing Semantic Reference Systems. In: Gould
MF, Laurini, R, Coulondre S (eds) 6
th
AGILE Conference on Geographic
Information Science, Presses Polytechniques et Universitaires Romandes,
April 24-26, 2003, Lyon, France, pp 63–72
Langacker RW (1987) Foundations of Cognitive Grammar, vol. 1: Theoretical
Prerequisites, Stanford University Press, Stanford
Mark DM (1993) Toward a Theoretical Framework for Geographic Entity Types.
In: Frank AU, Campari I (eds) Spatial Information Theory: Theoretical Basis
for GIS, Lecture Notes in Computer Science 716, Springer Heidelberg Berlin
New York, pp 270–283
Ogden CK, Richards IA (1923) The Meaning Of Meaning, Harcourt Brace
Jovanovich
Putnam H (1975) Mind, Language and Reality, Cambridge University Press,
Cambridge, MA
Quine WVO (1960) Word and Object, The MIT Press, Cambridge, MA
Scheider S, Janowicz K, Kuhn W (2009) Grounding Geographic Categories in the
Meaningful Environment, MUSIL working papers, Institute for Geoiformatics
(ifgi), University of Muenster, Muenster (Germany)
Smith B (2004) Beyond Concepts: Ontology as Reality Representation. In: Varzi
A, Vieu L (eds) Proceedings of FOIS
Smith B (2008) Ontology (Science). In: Eschenbach C, Grüninger, M (eds)
Formal Ontology in Information Systems (FOIS 2008), IOS Press
Sowa JF (2000) Knowledge Representation. Logical, Philosophical, and
Computational Foundations. Brooks Cole, Pacific Grove, CA
Vaníček P, Krakiwsky EJ (1982) Geodesy: The Concepts, North-Holland,
Amsterdam
Woods WA (1985) What’s in a link: Foundations for Semantic Networks. In:
Levesque RJBaHJ (ed) Readings in Knowledge Representation. Morgan
Kaufman, pp 218–241
Zadeh LA (2008) Is there a need for fuzzy logic? Information Sciences 178:
2751–2779