Content uploaded by Viviana Mascardi
Author content
All content in this area was uploaded by Viviana Mascardi
Content may be subject to copyright.
A Comparison of Upper Ontologies
Viviana Mascardi, Valentina Cordì
DISI, Università degli Studi di Genova,
Via Dodecaneso 35, 16146, Genova, Italy
E-mail: {cordi,mascardi}@disi.unige.it
Paolo Rosso
DSIC, Universidad Politécnica de Valencia,
Camino de Vera s/n, 46022, Valencia Spain
E-mail: prosso@dsic.upv.es
Abstract—Upper Ontologies are quickly becoming a key tech-
nology for integrating heterogeneous knowledge coming from
different sources. In fact, they may be exploited as a “lingua
franca” by intelligent software agents in all those scenarios where
it is impossible (or there is no will) for an agent to disclose its own
entire ontology to other agent, despite the need to communicate
with it. This paper represents the very preliminary step towards
the exploitation of Upper Ontologies as bridges for allowing
intelligent software agents to align heterogeneous ontologies in an
automatic way, where we analyse the most up-to-date state-of-the-
art. In this paper we analyse 7 Upper Ontologies, namely BFO,
Cyc, DOLCE, GFO, PROTON, Sowa’s ontology, and SUMO,
according to a set of standard software engineering criteria, and
we synthesise our analysis in form of a comparative table. A
summary of some existing comparisons drawn among subsets of
the 7 Upper Ontologies that we deal with in this document, is
also provided.
I. INTRODUCTION
The increasing pressing need that human and software
agents have to retrieve and exchange knowledge in a precise
and efficient way, have caused ontologies, web services, and
the combination of both, i.e., semantic web services, to be
more and more exploited for sharing knowledge within and
outside the boundaries of companies and other organisations.
Intelligent software agents are recognised by both researchers
and practitioners from the industry as one of the most suitable
means for mediating among the heterogeneity of applications
working within open, distributed, concurrent systems, and for
this reason they find application in many commercial projects.
However, there are still many unsolved issues for devel-
oping and deploying multi-agent systems (MASs) in open,
distributed, concurrent scenarios. One of them is how to find
mappings between concepts belonging to different ontologies
(in technical word, finding an alignment between these differ-
ent ontologies) in an automatic way. We are considering the
adoption of “Upper Ontologies” as bridges for making this
alignment possible.
Upper ontologies are quickly becoming a key technology
for integrating heterogeneous knowledge coming from differ-
ent sources. In fact, they may be used by different parties
involved in a knowledge integration and exchange process as
a reference, common model of the reality. In particular, they
may be exploited as a “lingua franca” by intelligent software
agents in all those scenarios where it is impossible (or there
is no will) for an agent to disclose its own entire ontology to
other agent, despite the need to communicate with it.
The definition of upper ontology (also named top-level
ontology, or foundation ontology) given by Wikipedia [22]
is “an attempt to create an ontology which describes very
general concepts that are the same across all domains. The
aim is to have a large number on ontologies accessible under
this upper ontology”.
This paper represents the very preliminary step towards
the exploitation of Upper Ontologies as bridges for allowing
intelligent software agents to align heterogeneous ontologies
in an automatic way, where we analyse the most up-to-date
state-of-the-art. In fact, in this paper we have described 7 upper
ontologies along different criteria that include dimension, im-
plementation language(s), modularity, developed applications,
alignment with the WordNet lexical resource, and licensing.
We have chosen these criteria for three reasons:
•They are software engineering criteria useful for the
developer of a knowledge-based system that has to choose
the most suitable Upper Ontology for his/her needs,
among a set of existing ones. Since all of us have a
computer science background, these criteria are more
familiar to us than philosophical ones.
•They take into account some of the evaluation questions
proposed by the IEEE Standard Upper Ontology Working
Group (http://suo.ieee.org/SUO/Evaluations/),
and they also extend the criteria considered in an existing
comparison among SUMO, Cyc, and DOLCE [18], thus
allowing us to “reuse”, and to be consistent with, the
results already obtained there.
•They are not (easily) scientifically falsifiable.
The choice of the 7 upper ontologies we have described,
namely BFO, Cyc, DOLCE, GFO, PROTON, Sowa’s ontology,
and SUMO, is based on how much, to the best of our
knowledge, they are visible and used inside the research
community. For example, we have discussed all the Upper
Ontologies referenced by Wikipedia, apart from WordNet that
we consider a lexical resource rather than an Upper Ontology,
and from the Global Justice XML Data Model and National
Information Exchange Model, that addresses the specific ap-
plication domain of justice and public safety. We have reported
alignments between the Upper Ontologies and WordNet, when
existing. To the 5 Upper Ontologies considered by Wikipedia,
we have added PROTON and Sowa’s ontology. We have
also cited three attempts to merge existing Upper Ontologies,
namely COSMO, MSO, and OntoMap, although we have not
described them in detail since the first two ones are still work
in progress, and the last one is over since four years.
The methodology followed to draw this paper consisted in
checking the existing literature, producing a first draft of the
comparison based on the retrieved literature, submitting it to
the attention of the developers of all the 7 upper ontologies
under comparison, and integrating the obtained answers and
suggestions into the current version of the paper. Due to time
constraints, we were not able to experiment with the upper
ontologies by our own. This “on the field” experimentation is
part of our near future work.
The paper is organised in the following way: Section II
provides a description of the 7 upper ontologies, and Section
III surveys some existing, partial comparisons drawn in the
past years among subsets of the Upper Ontologies that we
describe in Section II, and provides a synthesis of the results
of our comparison among them.
II. DESCRIPTION
a) Basic Formal Ontology (BFO):
•Status of this description. Validated by H. Stenzhorn,
research associate and doctoral student at the IFOMIS and
the University Hospital Freiburg - Medical Informatics
Department, and one of BFO’s developers.
•Home page. http://www.ifomis.org/bfo.
•Developers. B. Smith, P. Grenon, H. Stenzhorn, A. Spear
(IFOMIS, Saarland University).
•Description. BFO consists in two sub-ontologies: SNAP
– a series of snapshot ontologies (Oti), indexed by times
– and SPAN – a single videoscopic ontology (Ov). An
Oti is an inventory of all entities existing at a time, while
an Ovis an inventory of all processes unfolding through
time. Both types of ontology serve as basis for a series
of sub-ontologies, each of which can be conceived as a
window on a certain portion of reality at a given level of
granularity.
•History. The theory behind BFO has been developed and
formulated by Smith and Grenon in a series of publica-
tions starting in 1998. Its current implementation in OWL
has been developed by Stenzhorn with contributions from
Spear.
•Dimensions. BFO contains 1 top connecting class (“En-
tity”), 18 SNAP classes, and 17 SPAN classes for a
total of 36 classes which are, in version 1.0 of the
implementation, connected via the is_a relation. The
forthcoming version of BFO will incorporate relations
between classes too.
•Implementation language(s). OWL [21].
•Modularity. BFO is divided into the SNAP and SPAN
modules.
•Applications. BFO has been applied to the biomedical
domain [8] and it is currently used in building an ontology
for clinic-genomic trials on cancer (http://www.acgt-eu.
org).
•Alignment with WordNet. Not supported.
•Licensing. BFO is freely available; its OWL implemen-
tation may be downloaded from http://www.ifomis.org/bfo/
1.0.
b) Cyc:
•Status of this description. Validated by L. Lefkowitz,
executive director for business solutions at Cycorp.
•Home page. http://www.cyc.com/.
•Developers. Cycorp.
•Description. The Cyc Knowledge Base (KB) is a for-
malised representation of facts, rules of thumb, and
heuristics for reasoning about the objects and events of
everyday life. The KB consists of terms and assertions
which relate those terms. These assertions include both
simple ground assertions and rules. The Cyc KB is
divided into thousands of “microtheories” focused on
a particular domain of knowledge, a particular level of
detail, a particular interval in time, etc.
•History. The Cyc project was founded in 1984 by D.
Leant as a lead project in the Microelectronics and Com-
puter Technology Corporation (MCC). In 1994, Cycorp
was founded to further develop, commercialize, and apply
the Cyc technology. Cycorp offers a no-cost license to its
semantic technologies development toolkit to the research
community (ResearchCyc). Additionally, it has placed the
core Cyc ontology (OpenCyc) into the public domain.
•Dimensions. The Cyc KB (including Cyc’s microtheo-
ries) contains more than 300,000 concepts and nearly
3,000,000 assertions (facts and rules), using more than
15,000 relations.
•Implementation language(s). Cyc is represented in
the CycL formal language (http://www.cyc.com/cycdoc/ref/
cycl-syntax.html). The latest release of Cyc includes an
Ontology Exporter that allows to export specified portions
of Cyc to OWL files.
•Modularity. The “microtheory” approach supports mod-
ularity.
•Applications. Cyc has been used in the domains of
natural language processing, in particular for the tasks
of word sense disambiguation [4] and question answering
[5], of network risk assessment [19], and of representation
of terrorism-related knowledge [6].
•Alignment with WordNet. The last release of Cyc (as
well as of OpenCyc and ResearchCyc) includes links be-
tween Cyc concepts and about 12,000 WordNet synsets.
•Licensing. Cyc is a commercial product, but Cycorp
also released OpenCyc (http://www.opencyc.org/), the open
source version of the Cyc technology, and ResearchCyc
(http://research.cyc.com/), namely the Full Cyc ontology,
but with a research-only license.
c) DOLCE (a Descriptive Ontology for Linguistic and
Cognitive Engineering):
•Status of this description. Not validated by the ontology
developer(s).
•Home page. http://www.loa-cnr.it/DOLCE.html.
•Developers. Researchers from the Laboratory for Applied
Ontology, headed by N. Guarino.
•Description. DOLCE is the first module of the Won-
derWeb Foundational Ontologies Library. DOLCE has a
clear cognitive bias, in the sense that it aims at capturing
the ontological categories underlying natural language
and human commonsense. According to DOLCE, differ-
ent entities can be co-located in the same space-time.
DOLCE is described by its authors as an “ontology
of particulars”, which the authors explain as meaning
an ontology of instances, rather than an ontology of
universals or properties. The taxonomy of the most basic
categories of particulars assumed in DOLCE includes,
for example, abstract quality, abstract region, agentive
physical object, amount of matter, non-agentive physical
object, physical quality, physical region, process, tempo-
ral quality, temporal region.
•History. DOLCE has been developed as part of Wonder-
Web, a project funded as a shared-cost RTD under the
European Commission information society technologies
(IST) programme. WonderWeb started in 2002 and ended
in 2004. Although the project has already ended, DOLCE
is actively maintained and used.
•Dimensions. Around one hundred of terms, and a similar
number of axioms.
•Implementation language(s). First Order Logic, KIF [1],
OWL.
•Modularity. The intended use of DOLCE is within a
modular library of foundational ontologies, but it is not
currently divided into modules.
•Applications. According to the “DOLCE around the
world” web page (http://www.loa-cnr.it/dolcevar.html), there
are many projects that use DOLCE, including the LOIS
Project – an international research project on multilingual
information retrieval from legal databases –, SmartWeb
– a centre of excellence in research on intelligent
computing technologies and their application to web-
based systems and services –, Language Technology for
eLearning – a project funded by the EC, and using
multilingual language technology tools and semantic web
techniques for improving the retrieval of learning material
–, AsIsKnown – a semantic-based knowledge flow system
for the European home textiles industry, also funded by
the EC –, and the Projects of the Laboratory for Applied
Ontology.
•Alignment with WordNet. The OntoWordNet Project
aims at aligning the top-level of WordNet to DOLCE,
in order to obtain an “ontologically sweetened” lexical
resource, meant to be conceptually more rigorous, cog-
nitively transparent, and efficiently exploitable in several
applications. The beta version (v0.72) of the OWL align-
ment of WordNet 1.6 Noun Synsets to the DOLCE-Lite-
Plus ontology library consists of an alignment between
DOLCE-Lite-Plus and about 100 Wordnet sysnsets, and
can be downloaded from http://www.loa-cnr.it/ontologies/
OWN/OWN.owl.
•Licensing. The OWL version of DOLCE can be freely
downloaded from http://www.loa-cnr.it/ontologies/DLP3971.
zip.
d) GFO (General Formal Ontology):
•Status of this description. Validated by F. Loebe, PhD
student at the University of Leipzig under the supervision
of H. Herre and M. Löffler, members of the scientific
board of Onto-Med.
•Home page. http://www.onto-med.de/ontologies/gfo.html.
•Developers. The Onto-Med Research Group (http://www.
onto-med.de/).
•Description. GFO includes elaborations of categories like
objects, processes, time and space, properties, relations,
roles, functions, facts, and situations. Work is in progress
on an integration with the notion of levels of reality
in order to more appropriately capture entities in the
material, mental, and social areas.
•History. Work on GFO has started in 1999 in the context
of the GOL project (General Ontological Language).
Meanwhile, several directions of research have been
recognised and divided the initial project, such that GFO
is now one component of a larger framework. Work on
GFO remains in progress, because the development of
top-level ontologies is a long-term research effort.
•Dimensions. The OWL version of GFO consists of 79
classes, 97 subclass-relations, and 67 properties.
•Implementation language(s). The FOL axiomatization
of GFO and a KIF implementation of it are forthcoming.
An OWL-DL version also exists.
•Modularity. GFO exhibits a three-layered meta-
ontological architecture consisting of an abstract top
level, an abstract core level, and a basic level. The
foundational ontology GFO is structured into several
ontological modules including a module for functions and
a module for roles.
•Applications. One of the aims of the group Onto-Med
is the application of the GFO in the field of biomedical
science. GFO has been used to represent knowledge about
biological functions in the Gene Ontology, the Celltype
Ontology, and the Chemical Entities of Biological Interest
(ChEBI) Ontology [2], and GFO-Bio (http://onto.eva.mpg.
de/gfo-bio.html) is based on GFO and is a core ontology
for biology. Another area of application is the ontological
foundation of conceptual modelling. First examples of
applying GFO to UML are demonstrated in [9].
•Alignment with WordNet. Not supported.
•Licensing. The OWL version of GFO is released un-
der the modified BSD Licence (http://www.opensource.org/
licenses/bsd-license.php) and can be
downloaded from http://www.onto-med.de/ontologies/gfo.
owl.
e) PROTON (PROTo ONtology):
•Status of this description. Validated by A. Kiryakov,
head of Ontotext Lab, member of the board.
•Home page. http://proton.semanticweb.org/
•Developers. Ontotext Lab, Sirma (http://www.ontotext.
com/).
•Description. PROTON (PROTo ONtology) is a basic
upper-level ontology providing coverage of the general
concepts necessary for a wide range of tasks, including
semantic annotation, indexing, and retrieval of docu-
ments. The design principles can be summarized as
follows (i) domain-independence; (ii) light-weight logical
definitions; (iii) alignment with popular standards; (iv)
good coverage of named entities and concrete domains
(i.e. people, organizations, locations, numbers, dates, ad-
dresses).
•History. PROTON has been developed in the scope of
SEKT, a project co-funded by the EU 6th Framework pro-
gramme. SEKT started the 1st of January, 2004 and will
conclude at the end of 2006. PROTON is a development
of the KIMO ontology, which had been created and used
in the scope of the KIM platform for semantic annota-
tion, indexing, and retrieval (http://www.ontotext.com/kim/).
Currently, KIMO does not exist anymore; it is replaced
by PROTON, KIMLO (http://www.ontotext.com/kim/2005/
04/kimlo#) and KIMSO (http://www.ontotext.com/kim/2005/
04/kimso#).
•Dimensions. PROTON contains about 300 classes and
100 properties.
•Implementation language(s) A fragment of OWL Lite.
•Modularity. PROTON is organized in three levels includ-
ing four modules.
The System module ontology module occupies the first,
basic layer. It defines several notions and concepts of
a technical nature that are substantial for the operation
of any ontology-based software, such as semantic an-
notation and knowledge access tools. The Top ontol-
ogy module occupies the second layer and includes ba-
sic philosophically-reasoned distinctions between entity
types, such as Object, Happening, Abstract. Further up-
level, PROTON extends into its third layer, where either
of two independent ontologies, which defines much more
specific classes, can be used: PROTON Upper module or
PROTON KM (Knowledge Management) module. Exam-
ples of concepts belonging to these modules are Moun-
tain, as a specific type of Location, and ResourceCollec-
tion as a sub-class of InformationResource.
•Applications. As witnessed by a large number of pub-
lications (http://www.ontotext.com/publications/), PROTON
has been used in different domains and for different
purposes, including semantic annotation within the KIM
platform, and knowledge management systems in legal
and telecommunications domain [3]. It has also been used
as a basis for a domain ontologies in media research and
analysis (project MediaCampaign) and research intelli-
gence (project IST World), and a basis for Business Data
Ontology for Semantic Web Services [13].
•Alignment with WordNet. Not supported.
•Licensing. The four modules that compose PRO-
TON are freely accessible via Web: System module
(http://proton.semanticweb.org/2005/04/protons); Top module
(http://proton.semanticweb.org/2005/04/protont); Upper mod-
ule (http://proton.semanticweb.org/2005/04/protonu); Knowl-
edge Management module (http://proton.semanticweb.org/
2005/04/protonkm).
f) Sowa’s Ontology:
•Status of this description. Not validated by the ontology
developer(s).
•Home page. http://www.jfsowa.com/ontology/.
•Developers. J. F. Sowa.
•Description. Sowa’s ontology is based on [20]. The basic
categories and distinctions have been derived from a
variety of sources in logic, linguistics, philosophy, and
artificial intelligence. To keep the system open-ended,
Sowa’s ontology is not based on a fixed hierarchy of
categories, but on a framework of distinctions, from
which the hierarchy is generated automatically. For any
particular application, the categories are not defined by
drawing lines on a chart, but by selecting an appropriate
set of distinctions. These categories include Object, Pro-
cess, Schema, Script, Juncture, Participation, Description,
History, Structure, Situation, Reason, and Purpose. Each
of these categories may be either Physical or Abstract
(and in both cases, it can be either Continuant or Occur-
rent), and it may also be either Independent, Relative, or
Mediating. For example, Process is Physical, Occurrent
and Independent.
•History. Sowa’s ontology dates back to 1999. The two
major influences on it are the semiotics of C. Sanders
Peirce and the categories of existence of A. North White-
head.
•Dimensions. The KIF encoding of Sowa’s ontology con-
tains about 30 classes, 5 relationships among classes, and
among classes and instances (has, instance-of, subclass-
of, temp-part-of, spatial-part-of), about 30 axioms.
•Implementation language(s). Sowa’s ontology uses a
first-order modal language, i.e., a first-order language
with the modal operators “nec” and “poss”. A version
written in KIF also exists.
•Modularity. Sowa’s ontology is not explicitly divided
into modules, although each of the top level categories
can be intended as a module by its own, connected to the
other ones by means of relations.
•Applications. Sowa’s ontology inspired many existing
implemented upper ontologies, and thus its exploitation in
the development of “second-generation” upper ontologies
may be seen as one, and probably the most relevant, of
its practical applications.
•Alignment with WordNet. Not supported.
•Licensing. The KIF encoding of Sowa’s upper ontology
can be freely downloaded from http://suo.ieee.org/SUO/
ontologies/Sowa.txt.
g) SUMO (Suggested Upper Merged Ontology):
•Status of this description. Validated by A. Pease, current
Technical Editor of SUMO.
•Home page. http://www.ontologyportal.org/.
•Developers. The SUMO starter document was created at
Teknowledge by I. Niles and A. Pease, with a contribution
by C. Menzel.
•Description. SUMO and its domain ontologies [14] form
one of the largest formal public ontology in existence
today. They are being used for research and applications
in search, linguistics and reasoning. SUMO is extended
with many domain ontologies and a complete set of links
to WordNet, and is freely available.
•History. SUMO was first released in December 2000.
It was created at Teknowledge Corporation and it was
proposed as a starter document for the Standard Upper
Ontology Working Group (http://suo.ieee.org/), an IEEE-
sanctioned working group of collaborators from the
fields of engineering, philosophy, and information sci-
ence. SUMO was created by merging publicly available
ontological content into a single, comprehensive, and
cohesive structure. This content included the ontologies
available on the Ontolingua server (http://www.ksl.stanford.
edu/software/ontolingua/), Sowa’s upper level ontology, and
various mereotopological theories, among other sources.
•Dimensions. SUMO contains about 1000 terms and 4000
axioms; if we consider also the terms and axioms of its
domain ontologies, however, it reaches the dimension of
20,000 terms and 60,000 axioms.
•Implementation language(s). The first-order logic lan-
guage SUO-KIF (http://suo.ieee.org/SUO/KIF/suo-kif.html),
OWL.
•Modularity. SUMO consists of SUMO itself (the offi-
cial latest version on the IEEE web site can be down-
loaded from http://suo.ieee.org/SUO/SUMO/SUMO_173.kif),
the MId-Level Ontology (MILO), and ontologies of Com-
munications, Countries and Regions, Distributed Comput-
ing, Economy, Finance, Engineering Components, Ge-
ography, Government, Military, North American Indus-
trial Classification System, People, Physical Elements,
Transnational Issues, Transportation, Viruses, World Air-
ports. Additional ontologies of terrorism are available on
request.
•Applications. The applications of SUMO are docu-
mented by the almost one hundred published papers
describing its use (http://www.ontologyportal.org/Pubs.html).
The largest user community is in linguistics, but other
classes of applications include “pure” representation, and
reasoning. Applications range from academic to govern-
ment, to industrial ones.
•Alignment with WordNet. SUMO has been mapped
to all of Wordnet v2.1 by hand. The mappings can
be downloaded from http://sigmakee.cvs.source-
forge.net/sigmakee/KBs/WordNetMappings/.
•Licensing. SUMO is free and owned by the IEEE.
Its SUO-KIF implementation can be downloaded from
http://sigmakee.cvs.sourceforge.net/*check-
out*/sigmakee/KBs/Merge.kif, while the OWL
implementation can be downloaded from http:
//www.ontologyportal.org/translations/SUMO.owl.txt.
The ontologies that extend SUMO are available under
GNU General Public License.
h) Merging Upper Level Ontologies.: Three
attempts to merge some of the upper level
ontologies, thus leading to an “upper-upper level
ontology”, are COSMO (COmmon Semantic MOdel,
http://colab.cim3.net/cgi-bin/wiki.pl?CosmoWG/-
TopLevel), MSO (Multi-Source Ontology, http:
//www.webkb.org/doc/MSO.html), and the OntoMap Project
[11].
COSMO results from the efforts of the COSMO working
group (COSMO-WG) and its parent group, the Ontology
and Taxonomy Coordinating Working Group (ONTACWG).
COSMO is viewed as consisting of a lattice of ontologies
which will serve as a set of basic logically-specified concepts
(classes, relations, functions, instances) with which the mean-
ings of all terms and concepts in domain ontologies can be
specified. The use of a common set of defining concepts will
permit accurate interoperability of knowledge-based systems
using the logical relations of their ontologies as the basis
for reasoning in the system. Currently, COSMO integrates
concepts from the OpenCyc and SUMO ontologies, with some
classes from DOLCE and BFO. The work on COSMO is in
progress.
MSO is the Multi-Source Ontology of WebKB-2, a knowl-
edge server that permits Web users to browse and update
private knowledge bases on their machines, or alternatively,
a large shared knowledge base on the server machine. The
ontology of the shared knowledge base is currently an inte-
gration of various top-level ontologies and a lexical ontology
derived from an extension and correction of the noun-related
part of WordNet 1.7. The semantics of some categories from
WordNet has been modified in order to fix inconsistencies,
while the semantics of categories from other sources (e.g.
Sowa, DOLCE) has been kept. Also the work regarding the
MSO is still in progress. In particular, the integration of the
SUMO is still far from being complete. This integration links
the SUMO categories to the existing categories of the MSO,
adds some structure when needed, adds equivalent categories
the names of which are better suited for knowledge representa-
tion conventions that are “common” in the communities using
graph-based or frame-based notations, and finally translates the
axioms from KIF to more intuitive notations that permit people
to more easily understand the meanings of the categories and
their relationships.
Finally, OntoMap was a project with the goal to facilitate the
access, understanding, and reuse of such resources. A semantic
framework on conceptual level was implemented that was
small and easy enough to be learned on-the-fly. Technically,
OntoMap was implemented as a web-site providing access to
several upper-level ontologies and manual mapping between
them. OntoMap was similar in spirit to COSMO and MSO, but
only the very top concepts of each of the Upper Ontologies
considered there were aligned. Unfortunately, OntoMap was
over 4 years ago, and no maintenance was guaranteed to it. The
web-portal which was allowing online browsing is no longer
available, but the stand-alone viewer may be downloaded from
http://www.ontotext.com/projects/OntoMapViewer/install.htm.
III. COMPARISON
Some partial comparisons exist among subsets of the Upper
Ontologies that we have considered in Section II. In the next
paragraphs, we have summarised them in the most faithful
way. The interested reader should go to the source, always
cited, in order to have a complete picture of the conclusions
reached by the comparisons’ authors. The last paragraph,
instead, provides a synthesis of the description we have given
in Section II.
i) Pease’s comparison of DOLCE and SUMO.: In [15]
and [16], Pease draws a comparison between DOLCE and
SUMO. His conclusions are that DOLCE has a similar purpose
and business process to SUMO in that it is a free research
project for use in both natural language tasks and inference.
DOLCE has been carefully crafted with respect to strong
principles. DOLCE is an “ontology of particulars”; it does
have universals (classes and properties), but the claim is that
they are only employed in the service of describing particulars.
In contrast, SUMO could be described as an ontology of both
particulars and universals. It has a hierarchy of properties as
well as classes. This is a very important feature for practical
knowledge engineering, as it allows common features like
transitivity to be applied to a set of properties, with an
axiom that is written once and inherited by those properties,
rather than having to be rewritten, specific to each property.
Other differences include DOLCE’s use of a set of meta-
properties as a guiding methodology, as opposed to SUMO’s
use and formal definition of such meta-properties directly in
the ontology itself. With respect to SUMO, DOLCE does not
include such items as a hierarchy of process types, physical
objects, organisms, units and measures, and event roles.
j) Onto-Med’s comparison of GFO, DOLCE, and Sowa’s
ontology.: In [10], informal mappings from GFO to DOLCE
and from GFO to Sowa’s ontology, and viceversa, are spec-
ified. The authors of the comparison observe that all of
Sowa’s categories except for three can be reinterpreted in GFO.
However, mapping in the opposite direction seems to be more
problematic. For many of GFO categories, the corresponding
notions in Sowa’s ontology has not been found. Neither a
space-time model nor a property model is included in Sowa’s
ontology, and the construction method of GFO is not as strictly
combinatorial as is Sowa’s ontology. In DOLCE, levels of
reality are not introduced explicitly, while in GFO the authors
explicitly distinguish three levels of reality. Universals are
excluded from DOLCE, which supports neither the distinctions
provided in GFO concerning sets and items, nor concerning
the typology of categories. A time or a space model is
not built directly into DOLCE. Instead, the representation of
various models of space and time is permitted, which can be
introduced by means of qualities and their associated “qualia”
(the latter are similar to GFO’s quality values). In the GFO,
spatial location is modelled in terms of spatial regions and
relations, like occupation and location; temporal location is
based on time regions and projection relations. In addition,
presently the GFO provides a model for time and space. The
DOLCE distinction between endurant and perdurant is based
on the behavior of entities in time. Endurants are entities that
can change in time, are wholly present at any time of their
existence, and have no temporal parts but their parts are time-
indexed, and participate in perdurants. GFO distinguishes be-
tween persistence through time and being wholly present at a
time-boundary. This has produced two GFO categories instead
of endurant alone: persistants and presentials. GFO persistant
refers to the idea of persistence through time as attributed to
DOLCE’s endurant, although persistants are not considered in
GFO as individuals but as universals1. GFO presentials can
be generally interpreted as DOLCE endurants, but without
temporal extension. Intuitively, DOLCE notion of perdurant
corresponds to GFO notion of occurrent. Moreover, it seems
that the GFO notions of process, state and change can be
interpreted in DOLCE as stative, state and event, respectively.
Finally, the GFO categories that concern properties and their
values correspond rather well to DOLCE qualities, qualia and
quality spaces.
k) MITRE’s comparison of SUMO, Upper Cyc, and
DOLCE.: In [18], Semy, Pulvermacher and Obrst compare
SUMO, Upper Cyc, and DOLCE according to the existence
of an open license, modularity and evidence of use. We have
adopted these criteria inside our analysis, which thus subsumes
Semy, Pulvermacher and Obrst’s one.
l) Grenon’s comparison of DOLCE and BFO.: Grenon
made a careful comparison between DOLCE and BFO [7].
The conclusion is that both ontologies contain a category of
endurants and perdurants and an eternalist stance, and that the
theory of parthood and the theory of dependence are similar
in the two ontologies. Despite these similarities, there are also
many differences, including:
•DOLCE is methodologically fundamentally conceptualist
while BFO is methodologically fundamentally realist;
•DOLCE seems to be oriented toward commonsense, and
BFO’s naïve realism is in the same spirit. However,
DOLCE distinguishes between abstract and concrete en-
tities, and it includes agents and intentionality. BFO is
deliberately not committed to these distinctions. In par-
ticular, the physical / non-physical endurants distinction
in DOLCE is absent in BFO.
•As already mentioned, DOLCE is intended as an ontology
of particulars. BFO is intended to be an ontology of both
universals and particulars.
•In DOLCE, qualities are abstract entities which may not
be found in space or time, and do not have parts. For
BFO, the proxies of DOLCE’s qualities (“tropes”) are
located in space and exist at a time in the very same way
that the entities in which they inhere.
1The forthcoming release of GFO, expected by early 2007, will include
some refinements of the notion of persistence which will make this statement
no longer valid.
Another source of information about the similarities and
differences between DOLCE and BFO is [12], where Masolo,
Borgo, Gangemi, Guarino, and Oltramari of the Laboratory
For Applied Ontology (LOA) compare DOLCE and BFO
(besides the OCHRE object-centered ontology, [17], that we
did not consider in our analysis) by representing the assertion
“A statute of clay exists for a period of time going from t1to
t2. Between t2and t3, the statue is crashed and so ceases to
exists although the clay is still there.” in both of them.
m) Other existing sources of comparison.: Evaluations of
three Candidate Common Upper Ontologies, including SUMO
and MSO, can be found at http://suo.ieee.org/SUO/Evaluations/.
The criteria considered there include maturity, robustness,
potential for broad acceptance, language flexibility, owner-
ship/cost, and domain friendliness. These evaluations are not
comparative: each Upper Ontology is evaluated (usually, by
its creator) according to the above metrics.
n) Our comparison.: The description of the 7 Upper
Ontologies given in Section II is synthesised here in Tables I
and II.
IV. CONCLUSIONS
This paper represents a preliminary step towards the ex-
ploitation of upper ontologies as the means for allowing
intelligent software agents to integrate heterogeneous sources
of information, respecting privacy issues that are more and
more commong in many scenarios, such as virtual enterprises
and e-commerce. In fact, this paper provides an original and
unpublished analysis of the state-of-the-art in the field of upper
ontologies. This analysis is a necessary activity before starting
to think how upper ontologies may be actually exploited as a
bridge among two or more ontologies to be integrated. If the
original ontologies cannot be disclosed for privacy issues, each
agent involved in the application and responsible for accessing
and integrating one ontology, may “align” (i.e. find a mapping
between concepts) its own, private ontology, with the upper
ontology, and refer to the latter one in all its communicative
acts. At the time of writing, the design of an algorithm for
aligning ontologies using upper ontologies as a bridge is under
way. As soon as we will be able to implement and test it,
we will obtain results that will give us an important help
in understanding under which conditions the exploitation of
upper ontologies is feasible, and which upper ontologies are
better for being used as a bridge in the alignment process. Our
current and future work is entirely aimed at completing the
design and implementation of the algorithm and systematically
describing our experimental results.
ACK NOW LE DG ME NT S
We want to acknowledge all the researchers that helped in
drawing this comparison with their constructive comments and
useful advices. In particular, many thanks go to J. Euzenat, A.
Kiryakov, L. Lefkowitz, F. Loebe, A. Pease, J. Schoening, P.
Shvaiko, and H. Stenzhorn.
We also acknowledge the research projects TIN2006-15265-
C06-04 and “Iniziativa Software” CINI-FINMECCANICA
that partially funded this work.
REFERENCES
[1] American National Standard. KIF Knowledge Inter-
change Format – draft proposed American National Stan-
dard (dpANS) NCITS.T2/98-004, 1998.
[2] P. Burek, R. Hoehndorf, F. Loebe, J. Visagie, H. Herre,
and J. Kelso. A top-level ontology of functions and its
application in the open biomedical ontologies. In ISMB
(Supplement of Bioinformatics), volume 22, pages 66–73,
2006.
[3] N. Casellas, M. Blázquez, A. Kiryakov, P. Casanovas,
M. Poblet, and R. Benjamins. OPJK into PROTON:
Legal domain ontology integration into an upper-level
ontology. In R. Meersman and et al., editors, Proceed-
ings of the 3rd International Workshop on Regulatory
Ontologies (WORM 2005), volume 3762 of Lecture Notes
in Computer Science, pages 846–855. Springer, 2005.
[4] J. Curtis, D. Baxter, and J. Cabral. On the application
of the Cyc ontology to word sense disambiguation. In
Proceedings of the 19th International Florida Artificial
Intelligence Research Society Conference, pages 652–
657, 2006.
[5] J. Curtis, G. Matthews, and D. Baxter. On the effec-
tive use of Cyc in a question answering system. In
Proceedings of the IJCAI Workshop on Knowledge and
Reasoning for Answering Questions (KRAQ’05), 2005.
[6] C. Deaton, B. Shepard, C. Klein, C. Mayans, B. Sum-
mers, A. Brusseau, and M. Witbrock. The comprehensive
terrorism knowledge base in Cyc. In Proceedings of the
2005 International Conference on Intelligence Analysis,
2005.
[7] P. Grenon. BFO in a nutshell: A bi-categorial axiomati-
zation of BFO and comparison with DOLCE. Technical
Report 06/2003, IFOMIS, University of Leipzig, 2003.
[8] P. Grenon, B. Smith, and L. Goldberg. Biodynamic
ontology: Applying BFO in the biomedical domain. In
D. M. Pisanelli, editor, Ontologies in Medicine, volume
102 of Studies in Health Technology and Informatics,
pages 20–38. IOS Press, 2004.
[9] G. Guizzardi, H. Herre, and G. Wagner. On the general
ontological foundations of conceptual modeling. In
S. Spaccapietra, S. T. March, and Y. Kambayashi, editors,
Proceedings of the 21st International Conference on
Conceptual Modeling (ER 2002), volume 2503 of Lecture
Notes in Computer Science, pages 65–78. Springer, 2002.
[10] H. Herre, B. Heller, P. Burek, R. Hoehndorf, F. Loebe,
and H. Michalek. General formal ontology (GFO) –
part I: Basic principles. Technical Report 8, Onto-Med,
University of Leipzig, 2006.
[11] A. Kiryakov, K. Ivanov Simov, and M. Dimitrov. On-
toMap: portal for upper-level ontologies. In Proceedings
of the 2nd International Conference on Formal Ontology
in Information Systems (FOIS 2001), pages 47–58. ACM,
2001.
[12] C. Masolo, S. Borgo, A. Gangemi, N. Guarino, and
A. Oltramari. Ontology library (final). IST Project 2001-
33052 WonderWeb Deliverable D18, 2003.
[13] G. Nagypál and J. Lemcke. A business data ontology.
Data, Information and Process Integration with Semantic
Web Services Project, FP6 ˝
U 507483, Deliverable D3.3,
2005.
[14] I. Niles and A. Pease. Towards a standard upper on-
tology. In C. Welty and B. Smith, editors, Proceedings
of the 2nd International Conference on Formal Ontology
in Information Systems (FOIS-2001), pages 2–9. ACM
Press, 2001.
[15] A. Pease. Formal representation of concepts: The Sug-
gested Upper Merged Ontology and its use in linguistics.
In A. C. Schalley and D. Zaefferer, editors, Ontolin-
guistics. How Ontological Status Shapes the Linguistic
Coding of Concepts. Mouton de Gruyter, 2006.
[16] A. Pease and C. Fellbaum. Formal ontology as inter-
lingua: The SUMO and WordNet linking project and
GlobalWordNet. To appear.
[17] L. Schneider. How to build a foundational ontology: The
object-centered high-level reference ontology OCHRE.
In A. Günter, R. Kruse, and B. Neumann, editors,
Proceedings of the 26th Annual German Conference on
AI, KI 2003: Advances in Artificial Intelligence, volume
2821 of Lecture Notes in Computer Science, pages 120–
134. Springer, 2003.
[18] S. K. Semy, M. K. Pulvermacher, and L. J. Obrst. Toward
the use of an upper ontology for U.S. government and
U.S. military domains: An evaluation. Technical Report
MTR 04B0000063, The MITRE Corporation, 2004.
[19] B. Shepard, C. Matuszek, C. B. Fraser, W. Wechten-
hiser, D. Crabbe, Z. Güngördü, J. Jantos, T. Hughes,
L. Lefkowitz, M. J. Witbrock, D. B. Lenat, and E. Larson.
A knowledge-based approach to network security: Ap-
plying Cyc in the domain of network risk assessment. In
M. M. Veloso and S. Kambhampati, editors, Proceedings
of the 20th National Conference on Artificial Intelligence
and the 17th Innovative Applications of Artificial Intelli-
gence Conference, pages 1563–1568. AAAI Press AAAI
Press / The MIT Press, 2005.
[20] J. F. Sowa. In Knowledge Representation: Logical,
Philosophical, and Computational Foundations. Brooks
Cole Publishing, 1999.
[21] W3C. OWL Web Ontology Language Overview – W3C
Recommendation 10 February 2004 , 2004.
[22] Wikipedia. Upper ontology – wikipedia, the free ency-
clopedia, 2006. [Online; accessed 15-December-2006].
Home page Developers Dimensions Language(s)
BFO http://www.
ifomis.org/bfo
Smith, Grenon,
Stenzhorn, Spear
(IFOMIS)
36 classes related via the
is_a relation OWL
Cyc http://www.cyc.
com/ Cycorp
About 300,000 concepts,
3,000,000 assertions (facts
and rules), 15,000 rela-
tions (these numbers in-
clude microtheories)
CycL, OWL
DOLCE
http://www.
loa-cnr.it/
DOLCE.html
Guarino and other
researchers of the
LOA
About 100 concepts and
100 axioms
First Order
Logic, KIF,
OWL
GFO
http://www.
onto-med.de/
ontologies/gfo.
html
The Onto-Med Re-
search Group
79 classes, 97 subclass-
relations, 67 properties
First Order
Logic and
KIF (forth-
coming);
OWL
PROTON http://proton.
semanticweb.org/
Ontotext Lab,
Sirma
300 concepts and 100
properties OWL Lite
Sowa’s
http://www.
jfsowa.com/
ontology/
Sowa 30 classes, 5 relationships,
30 axioms
First Order
Modal
Language,
KIF
SUMO
http://www.
ontologyportal.
org/
Niles, Pease, and
Menzel
20,000 terms and 60,000
axioms (including domain
ontologies)
SUO-KIF,
OWL
Table I
COMPARISON, PART I
Modularity Applications Alignment with
WordNet Licensing
BFO
SNAP
and SPAN
modules
Mainly in the biomedical do-
main Not supported Freely available
Cyc “Microtheory”
modules
Natural language processing,
network risk assessment, ter-
rorism management
Cyc is mapped
to about 12,000
WordNet synsets
Commercial
product;
ResearchCyc
and OpenCyc
are instead
freely available
(ResearchCyc for
research purposes
only)
DOLCE Not divided
into modules
Multilingual information re-
trieval, web-based systems
and services, e-learning
DOLCE-Lite-Plus
has been aligned
with about 100
Wordnet sysnsets
Freely available
GFO
Abstract
top level,
abstract core
level, basic
level
Mainly in the biomedical do-
main Not supported
Released under the
modified BSD Li-
cence
PROTON
Three levels
including
four modules
Semantic annotation within
the KIM platform, knowl-
edge management systems
in legal and telecommunica-
tions domain, media research
and analysis, research intelli-
gence, Business Data Ontol-
ogy for Semantic Web Ser-
vices.
Not supported Freely available
Sowa’s Not divided
into modules
No documented applications
have been developed, but
Sowa’s ontology inspired the
creation of many imple-
mented Upper Ontologies
Not supported Freely available
SUMO
Divided
into SUMO
itself, MILO,
and domain
ontologies
Linguistics, representation,
reasoning
SUMO has been
mapped to all of
Wordnet v2.1 by
hand
Freely available
Table II
COMPARISON, PART II
