=Paper=
{{Paper
|id=Vol-1495/paper_15
|storemode=property
|title=Enhancing Terminological Knowledge with Upper Level Ontologies
|pdfUrl=https://ceur-ws.org/Vol-1495/paper_15.pdf
|volume=Vol-1495
|dblpUrl=https://dblp.org/rec/conf/tia/SeppalaH15
}}
==Enhancing Terminological Knowledge with Upper Level Ontologies==
https://ceur-ws.org/Vol-1495/paper_15.pdf
Proceedings of the conference Terminology and Artificial Intelligence 2015 (Granada, Spain)
179
Enhancing Terminological Knowledge With Upper Level Ontologies
Selja Seppälä Amanda Hicks
University at Buffalo University of Florida
Buffalo, NY, USA Gainesville, FL, USA
seljamar@buffalo.edu aehicks@ufl.edu
Abstract the same natural language or in different ones.
This definition may also lead to a multiplication
In this communication, we advocate the use of ontological terms expressing categories and re-
of upper level ontologies such as the Basic
lations to represent the same or distinct conceptual
Formal Ontology (BFO) to enhance termi-
nological resources and research. First, we systems.
present common issues in ontologized ter- However, a multiplication of ontological meta-
minological work. Then, we review two languages (categories and relations) tends to cre-
projects that illustrate the potential advan- ate knowledge silos (Smith and Ceusters, 2010).
tages of integrating rigorous formal upper In particular, when these metalanguages are
level ontologies. Finally, we discuss pos- domain-specific. Even within a single domain, us-
sible challenges and conclude with a sum-
ing distinct metalanguages can limit interoperabil-
mary of the benefits that such ontologies
can bring to both terminological theory and ity of systems using ontological representations
practice. of terminologies. Furthermore, from the termi-
nological research viewpoint, a multiplication of
categories and relations hinders the advancement
1 Introduction
of our understanding of conceptual systems, of
Terminologies encode lexical and background the internal structure of terms and definitions, etc.
knowledge that experts have about their domain To avoid these limitations, we propose that termi-
of expertise. These resources can be associ- nologists developing terminological resources and
ated with a more explicit ontology-like represen- carrying out research would greatly benefit from
tation of the entities in the relevant domain. Such using an upper level ontology, such as the Ba-
representations may include, for example, non- sic Formal Ontology (BFO), to integrate resources
lexicalized concepts. This extends mere termi- and research.
nologies to more sophisticated knowledge repre- In this communication, we present and discuss
sentations. Being language independent, ontolo- existing works integrating upper level ontologies,
gized terminologies have the advantage of inte- and underline the main advantages of augmenting
grating multilingual terminologies. When aug- terminological knowledge with categories and re-
mented with axioms, they can be used in reasoning lations from an upper level ontology such as BFO.
systems.
Terminological works, where they refer to on- 2 Limitations of Ontological
tologies at all, generally use Gruber’s definition of Terminologies
an ontology as “an explicit specification of a con- As shown in Seppälä (2012), common issues in
ceptualization.” (Gruber, 1995). Ontologies built ontologized terminologies are:
on the basis of this definition thus depend on peo-
ples’ concepts. As a result, the Gruber approach • Lack of rigorously defined categories and re-
may lead to several distinct ontological represen- lations. The interpretation of the metalan-
tations of the same domain, whether expressed in guage is left to our intuitive understanding of
Proceedings of the conference Terminology and Artificial Intelligence 2015 (Granada, Spain)
180
the terms used for expressing the used cate- 3 Enhancing Terminologies with Upper
gories and relations. Level Ontologies
A formal upper level ontology can be defined as
• is a overloading (Guarino, 1998): the is a
“a representation of the categories of objects and
relation used for structuring the domain on-
of the relationships within and amongst categories
tology does not distinguish the genuine is a
that are to be found in any domain of reality what-
subsumption relation from the instance of
soever.” (Spear, 2006)
relation, and sometimes even from the
To illustrate the potential advantages for termi-
part of relation.
nology of using formal upper level ontologies, we
describe two projects that integrate such ontolo-
• Multiplication of domain-specific, some-
gies. There are a few upper level ontologies that
times ad hoc, categories and relations.
can be used by mid-level or domain-specific on-
tologies to define and relate their categories in a
• When upper level categories are used, limita-
non-ambiguous manner, using logical axioms if
tion to a few top-most categories, which are
needed. The projects described hereafter use, re-
completed with domain-specific ones (Faber,
spectively, the Descriptive Ontology for Linguistic
2002; Kageura, 2002).
and Cognitive Engineering (DOLCE) (Masolo et
al., 2001) and the Basic Formal Ontology (BFO)
The above limitations result in practical and (Arp et al., 2015).
research-related consequences for terminological
works, which can be summarized as follows: 3.1 The KYOTO Project
The KYOTO project aims at representing domain-
• Confusing and incompatible representations specific terms in a computer-tractable axiomatized
of the same domain. formalism to allow machines to reason over texts
in natural language (Vossen et al., 2010). The sys-
• Non-interoperable terminologies, which hin- tem developed in this project comprises a platform
ders the possibility of sharing and reusing ter- for multilingual text mining and information ex-
minological resources. traction that was tested on documents from the en-
vironmental domain. The semantics of the terms
• Non-generalizable observations of termino- are defined through the KYOTO ontology which
logical phenomena, which hinders research is based on DOLCE. WordNets and specialized
towards a proper understanding of content- vocabularies of different languages are linked to
related principles governing term formation, ontology classes on the basis of a mapping of the
definition composition, and conceptual sys- English WordNet to the KYOTO ontology. “This
tem organization. This eventually hinders the basic ontology and the mapping to WordNet are
development of widely (re)usable termino- used to model the shared and language-neutral
logical tools, for example, for creating new concepts and relations in the domain.” (Vossen et
terms and writing definitions. al., 2010, 4) The system can thus “detect similar
data across documents in different languages, even
• Non-comparable results of terminological re- if expressed differently.” (Vossen et al., 2010, 2)
search for lack of a common well-defined In Vossen et al. (2013), the authors extracted
domain- and language-independent metalan- statements from texts about the Chesapeake Bay
guage, which hinders the development of a using Kybots, scripts based on ontological and lin-
mature integrated science. guistic patterns in annotated text. The results of
baseline fact extraction were compared with Ky-
These shortcomings can be addressed by bot extraction and Cterm extraction, both of which
adopting well-defined domain- and language- utilize the KYOTO ontology. The result was that
independent upper level categories and relations the baseline and Kybot profiles had high recall,
(ontological metalanguage) of the sort accounted 100% and 91% respectively. The baseline had
for in formal upper level ontologies. low precision (18%), whereas the precision of the
Proceedings of the conference Terminology and Artificial Intelligence 2015 (Granada, Spain)
181
Kybot profiles was better, though not optimal, at In those cases, it may not be straightforward
31%. In short, leveraging ontological information under which category to place a term.
in domain-specific fact extraction NLP resulted in
high recall and improved precision. • Specifications of the upper level ontology
may be sparse and lacking, and sometimes
3.2 The BFO-Based Ontological Analysis too formal (OWL, first order logic) to be eas-
Framework ily understood by terminologists.
The second project consists in analyzing the con-
• An adequate use requires familiarity with the
tents of definitions using the categories and rela-
upper level ontology chosen.
tions of BFO (Seppälä, 2012; Seppälä, 2015b).
The author puts forward an ontological anal- A solution to such issues would be to use exist-
ysis framework that is domain- and language- ing mappings of WordNet to upper level ontolo-
independent and that can be used in any kind of gies as aids for integration. A future mapping
terminological conceptual analysis task. The cat- of WordNet to BFO should facilitate the integra-
egories and their characteristics are also used as tion of BFO in terminological projects (Seppälä,
models that serve to predict the contents of defini- 2015a).
tions. These may be used as templates in tools to
help in definition writing. 5 Conclusion
The results of the pilot study reported in
Seppälä (2012; 2015b) show that these BFO- We saw that ontologized terminologies present a
Templates account for about 75% of the contents number of shortcomings that can be addressed by
of definitions of terms from 15 distinct domains. integrating a formal upper level ontology. We il-
The rest of the definition contents can be described lustrated the advantages of such an enhancement
using the BFO categories and relations. by reviewing two projects that use such ontolo-
gies. To summarize, the main benefits of using
The well-defined BFO vocabulary can thus be
a language- and domain-independent upper level
used as a metalanguage to describe definition con-
ontology are, on the practical side, the possibility
tents, term formation, and the organization of
to integrate multilingual and multi-domain termi-
conceptual systems in a way that research find-
nological resources with one another and with in-
ings can be compared and integrated. In prac-
formation system tools. The latter can thus use the
tice, BFO-based ontologized terminologies would
inferences drawn on the basis of the upper level
have the advantage of being interoperable, as it
ontology to reason over and manipulate multilin-
is already the case for the mid-level and domain-
gual natural language texts. Using a well-defined
specific ontologies (and the corresponding termi-
formal upper level ontology as a basis for termino-
nologies) that extend BFO, such as the Ontology
logical work would make sharing and reuse of ter-
for Biomedical Investigations (OBI) and the On-
minologies easier: identifying and sharing com-
tology for Biobanking (OBIB)1 .
mon terms, constructing new definitions using the
4 Possible Obstacles to Use of Upper same building blocks (information types and logi-
Level Ontologies cal axioms), etc. Such a framework avoids seman-
tic conflicts and need for mapping.
Using upper level ontologies may sometimes On the research side, using a well-defined on-
prove challenging. Possible issues may be: tological metalanguage allows: carrying out rig-
orous and comparable conceptual analysis work
• Upper level ontologies evolve and their cate- in terminology; making language- and domain-
gories are, at times, still under development. independent generalizations about term formation,
1
For a full list, see http://ifomis. definition content structure, and terminological
uni-saarland.de/bfo/users. For an illus- systems’ organization, which can help develop
tration of interoperability and its advantages, see the empirically based content standards and writing
presentation on The OBIB Ontology for Biobanking, by
Chris Stoeckert, Jie Zheng, and Mathias Brochhausen
aid tools; creating comparable research results that
http://ncorwiki.buffalo.edu/index.php/ contribute to developing a mature integrated ter-
CTS_Ontology_Workshop_2015. minological science.
Proceedings of the conference Terminology and Artificial Intelligence 2015 (Granada, Spain)
182
Moreover, a metalanguage using the categories International Journal of Human Computer Studies,
and relations of an upper level ontology for de- 43(5):907–928.
scribing terminological data (for example, terms’, Nicola Guarino. 1998. Some ontological principles
definitions’, and conceptual systems’ structure) for designing upper level lexical resources. In Pro-
ceedings of First International Conference on Lan-
can fruitfully complement any terminological re-
guage Resources and Evaluation. ELRA-European
source whether or not already ontologized. Cimi- Language Resources Association, Granada, Spain,
ano et al. (2011) propose, for example, a model pages 527–534. Citeseer.
to formally link lexicons (with relevant linguistic Kyo Kageura. 2002. The Dynamics of Terminology:
descriptions) to ontologies. A Descriptive Theory of Term Formation and Termi-
Using more specifically a BFO-based metalan- nological Growth. Terminology and Lexicography
guage would further enhance our understanding Research and Practice 5. John Benjamins, Amster-
of the relationship between the lexical, linguis- dam.
tic, conceptual, and ontological levels of termi- Claudio Masolo, Stefano Borgo, Aldo Gangemi,
Nicola Guarino, and Alessandro Oltramari. 2001.
nologies. Indeed, BFO is a realist ontology that Wonderweb deliverable D18 ontology library (fi-
represents the things that exist in the world and nal). ICT Project.
the relations between them, independently of our Selja Seppälä. 2012. Contraintes sur la
conceptualizations thereof. A BFO-based meta- sélection des informations dans les définitions ter-
language may thus provide an additional level of minographiques: vers des modèles relationnels
understanding to existing descriptive frameworks. génériques pertinents. Ph.D. thesis, Département
We therefore encourage terminologists to fully de traitement informatique multilingue (TIM), Fac-
ulté de traduction et d’interprétation, Université de
embrace the best ontological practices to enhance
Genève.
their research and resources.
Selja Seppälä. 2015a. Mapping WordNet to the Basic
Formal Ontology using the KYOTO ontology. In
Acknowledgments
Proceedings of ICBO 2015.
This work was supported in part by the Swiss Selja Seppälä. 2015b. An ontological framework for
National Science Foundation (SNSF) and by modeling the contents of definitions. Terminology,
21(1):23–50.
the NIH/NCATS Clinical and Translational Sci-
Barry Smith and Werner Ceusters. 2010. Ontological
ence Awards to the University of Florida UL1
Realism: A Methodology for Coordinated Evolution
TR000064. The content is solely the responsibil- of Scientific Ontologies. Applied Ontology, 5:139–
ity of the authors and does not necessarily repre- 188.
sent the official views of the SNSF, the National Andrew D. Spear, 2006. Ontology for the Twenty First
Institutes of Health, or the NCTE. Thanks also to Century: An Introduction with Recommendations.
Aurélie Picton and Barry Smith. Institute for Formal Ontology and Medical Informa-
tion Science, Saarbrücken, Germany.
Piek Vossen, German Rigau, Eneko Agirre, Aitor
References Soroa, Monica Monachini, and Roberto Bartolini.
2010. KYOTO: an open platform for mining facts.
Robert Arp, Barry Smith, and Andrew D. Spear. 2015.
In Proceedings of the 6th Workshop on Ontologies
Building Ontologies with Basic Formal Ontology.
and Lexical Resources, pages 1–10.
MIT Press, Cambridge, MA.
Piek Vossen, Eneko Agirre, German Rigau, and Aitor
Philipp Cimiano, Paul Buitelaar, John McCrae, and
Soroa. 2013. Kyoto: A knowledge-rich approach
Michael Sintek. 2011. LexInfo: A declarative
to the interoperable mining of events from text. In
model for the lexicon-ontology interface. Web Se-
New Trends of Research in Ontologies and Lexical
mantics: Science, Services and Agents on the World
Resources, pages 65–90. Springer.
Wide Web, 9(1):29–51.
Pamela Faber. 2002. Terminographic definition
and concept representation. In Johann Haller Be-
linda Maia and Margherita Ulyrich, editors, Train-
ing the Language Services Provider for the New Mil-
lennium, pages 1–14 [343–354]. Universidade do
Porto, Porto.
Thomas R. Gruber. 1995. Toward Principles for the
Design of Ontologies Used for Knowledge Sharing.