B y G r a e m e S h a n k s , E l i z a b e t h Ta n s l e y ,

a n d R o n Weber

USING ONTOLOGY
TO VALIDATE CONCEPTUAL MODELS

conceptual model is a representation

A
(typically graphical) constructed by IS
professionals of someone’s or some
group’s perception of a real-world
domain. It might be used to facilitate
the design and implementation of an Theories of ontology
information system. It might be used to lead to improved
evaluate the fit between an organization’s needs conceptual models
and the business models embedded within an and help ensure they
enterprise application software package.
After constructing a conceptual model, IS are indeed faithful
professionals need to validate it with the stakehold- representations of
ers whose worlds they are seeking to represent. their focal domains.
Otherwise, defects in the model might propagate
to subsequent system design and implementation
activities. If these defects are not discovered
until late in the development process, they are
often costly to correct. Validating a conceptual
model is thus critical to high-quality system
development.

COMMUNICATIONS OF THE ACM October 2003/Vol. 46, No. 10 85

 Validating a conceptual model involves checking design-and-code inspections of programs, IS profes-
that it faithfully represents the domain it is intended sionals and users who had no part in the preparation
to represent (the focal domain). It is faithful if it has of a conceptual model may still participate in its vali-
certain attributes: dation.
Methodology. There are several approaches:
Accuracy. The model should accurately represent
the semantics of the focal domain as perceived by Review. Participants can be asked to review and
the focal stakeholder(s); evaluate the model in whatever way they choose;
Completeness. The model should completely repre- Questioning. Participants can ask one another about
sent the semantics of the focal domain as per- the domain being modeled based on the content
ceived by the focal stakeholder(s); of the model;
Conflict-free. The semantics represented in different Problem solving. Using the conceptual model, par-
parts of the model should not contradict one ticipants can be asked to solve problems about
another; and the focal domain. They might reflect the scenar-
No redundancy. To reduce the likelihood of con- ios or use-cases sometimes employed in OO sys-
flicts arising if and when the model is subse- tems development; and
quently updated the model should not contain Transaction testing. Events in the focal domain
redundant semantics. (transactions) can be evaluated to determine
whether the things that experience the events are
Unfortunately, little is known about how to validate faithfully represented in the conceptual model.
conceptual models effectively and efficiently.
Although [5], for example, provides more useful Developing conceptual models that facilitate valida-
guidance than most sources, recommending that tion processes is certaintly difficult. IS professionals
IS professionals follow two validation tasks— often struggle to find effective ways to get stake-
test transactions against a conceptual model and holders to engage meaningfully with the conceptual
review the conceptual model with users—scant models they have produced.
specifics are provided about how these tasks should
be carried out. Ontologies
Can stakeholders improve the validation of concep-
Approaches to Validation tual models by using formal ontologies (theories
The question of how to conduct cost-effective vali- about the structure and behavior of the real world in
dations of conceptual models is like the question of general), as opposed to domain ontologies (such as
how to conduct cost-effective validation of pro- those pertaining to medicine) and task ontologies
grams. Three issues need to be addressed: (such as those pertaining to vascular surgery)?
Scope. Only a relatively small subset of a conceptual Ontologies are unlikely to help with decisions about
model might be validated, perhaps focusing on just scope and participation in the validation process.
one focal stakeholder’s views of some part of an appli- However, in terms of validation methodology, they
cation domain (such as unit testing of a program). help in three important ways: choosing the concep-
Alternatively, a larger subset of the conceptual model tual modeling grammar for representing the focal
might be validated, perhaps focusing on several focal domain; understanding the phenomena represented
stakeholders’ views of different but nonetheless over- in conceptual modeling scripts (diagrams); and
lapping parts of an application domain (such as inte- making sense of ambiguous semantics in conceptual
gration testing of a program). Or the entire models.
conceptual model might be validated (such as whole- Choosing a conceptual modeling grammar. Concep-
of program testing of a program). Moreover, as more tual models are expressed using conceptual modeling
and more interorganizational systems are imple- grammars that provide constructs for representing
mented, increasingly conceptual models spanning real-world phenomena and rules defining how these
organizations must be validated. constructs might be combined to represent focal
The people involved in the process. Participants might domains. For example, the original entity-relation-
include the focal stakeholders and IS professionals ship (ER) modeling grammar provided three con-
who developed the conceptual model, as well as indi- structs for representing phenomena in the world:
viduals independent of the development of the con- entity (and entity set); relationship (and relationship
ceptual model. Just as different people take part in set); and attributes [4]. Using them to build concep-

86 October 2003/Vol. 46, No. 10 COMMUNICATIONS OF THE ACM

tual models of a domain is governed by certain rules;
for example, two or more entities have to be con-
nected via a relationship.
Different conceptual modeling grammars reflect
different strengths and weaknesses in terms of their
ability to generate scripts (diagrams) to represent dif-
ferent types of phenomena that might occur in a
domain; for example, the ER modeling grammar’s
strengths derive from its ability to represent static phe-
nomena rather than dynamic phenomena in a
domain. At the outset, stakeholders’ ability to validate
a conceptual model would be undermined if the con-
ceptual modeling grammar they use to generate
descriptions of the domain provides only weak repre-
sentations of key phenomena in the domain. UNFORTUNATELY, LITTLE
Theories of ontology can help structure stakehold- IS KNOWN ABOUT
ers’ identification of the different types of phenomena
in a domain; for instance, using Bunge’s theory of HOW TO VALIDATE
ontology [2], phenomena might be classified as CONCEPTUAL MODELS
things, properties of things, states of things, laws,
events in things, or couplings. Some types of phe- EFFECTIVELY AND
nomena in the focal domain are likely to be more
common or more important than others. Stakehold- EFFICIENTLY.
ers must therefore choose conceptual modeling gram-
mars that faithfully represent these key phenomena.
Theories of ontology can also be used to evaluate
the strengths and weaknesses of competing concep-
tual modeling grammars as a means of representing
the key phenomena in the focal domain. They can be
used to evaluate the completeness of a conceptual
modeling grammar, that is, the extent to which the
grammar provides constructs useful in representing
the different types of phenomena in the focal domain
[11]; for example, using the grammar, would it be
possible to generate a representation of the lawful state
space of a thing in the focal domain? If certain types
of phenomena cannot be represented via a particular
grammar, theories of ontology can be used to identify
other grammars that would cover these phenomena.
In this way, IS professionals can astutely determine
how to use multiple grammars in concert.
Theories of ontology can also be used to evaluate the
clarity of a conceptual modeling grammar, that is, the
extent to which a one-to-one mapping exists between
different types of phenomena and modeling constructs
provided in the grammar [11]; for instance, using the
grammar, would construct overload exist because a
thing and an event both have to be represented by the
same grammatical construct (such as an entity symbol)?
When the grammar is not clear, steps might be taken to
reduce ambiguities; for instance, overloaded constructs
might be annotated in different ways when used to rep-
resent different types of phenomena.

COMMUNICATIONS OF THE ACM October 2003/Vol. 46, No. 10 87

 The overall goal is to choose a conceptual model- Ontologists distinguish among different types of
ing grammar (or set of grammars) that is complete phenomena in their theories because they view the
and clear in terms of representing the key phenomena various distinctions as real and thus important. Using
in the focal domain. an ontological theory to identify misclassified phe-
Representing specific domain phenomena. A concep- nomena in a conceptual model can assist stakeholders
tual modeling grammar that is well-suited to model- in validating the model. The reason is that the way
ing a domain might still be used inappropriately. The some phenomenon is classified is likely to prompt cer-
resulting poor-quality models undermine stakehold- tain questions about the phenomenon and not other
ers’ ability to validate them. questions about the same phenomenon; for instance,
Theories of ontology can be used to inform con- if a particular phenomenon is classified as an entity,
ceptual modelers about how to represent phenomena stakeholders are likely to focus on ensuring the attrib-
in the focal domain. They provide a foundation for utes of this entity are specified completely and accu-
modeling rules that indicate how the constructs pro- rately in the conceptual model. If a particular
vided in conceptual modeling grammars should be phenomenon is classified as a law, however, these same
used to model phenomena in the focal domain; for stakeholders are likely to focus on ensuring the lawful
example, based on Bunge’s state space and lawful
theory of ontology, prior Aircraft
Type
in Landing
Rule
for Airport
Type
event space associated
research has argued that five classifies
with the entities covered
rules should be adopted by by the law are specified
Aircraft classifies
IS professionals when mod- correctly and completely.
eling a domain: scheduled on
scheduled
The representation
Flight
as
Airline arrival problem with the figure is
Route Airport
• Composites and aggre- for departure
that it motivates stake-
gates should be modeled operated by holders to ask validation
Ticket
as entities, not as rela- Airline
questions about the focal
tionships [10]; travels on domain as though all phe-
• Relationships should not Passenger groups Passenger nomena in it were either
be modeled with attrib- Category entities or relationships; for
utes [3]; instance, because “landing
• Entities should not be modeled with optional Normalized entity-relationship rule” is represented as an
attributes [1, 10]; model for airline ticketing and entity rather than a law,
scheduling.
• Conceptual models should clearly distinguish systems analysts involved
between classes and instances [6, 7]; and in the validation process
• Things and their properties should be clearly dis- might be limited to their understanding of the model
tinguished in conceptual models [8]. if they take a relational-table (implementation) view of
the rule. The questions they ask of end users might
However, some of these rules contradict widely used focus exclusively on eliciting valid aircraft-type/airport-
practices in conceptual modeling. They have been type pairs. As a result, they could fail to tease out sub-
developed to impose discipline on conceptual mod- tle yet critical aspects of landing rules that cannot be
eling practices that lead to conceptual models that represented easily as relational tables but nonetheless
are faithful representations of the focal domain and constrain the state spaces and event spaces of aircraft
more easily understood by stakeholders. and airports.
Making sense of ambiguous semantics. When stake- The dual classification of phenomena, shown in
holders have to validate a conceptual model, they the figure as entities and relationships, is likely to
sometimes use theories of ontology to help clarify prompt only certain kinds of questions when validat-
ambiguous semantics in the model; for example, in ing the conceptual model. Consequently, validation
the figure here, which outlines a conceptual model of the semantics of phenomena in the focal domain
typical of many used in practice [9], all phenomena in that are not entities or relationships might be done
the domain are represented as either entities or rela- poorly because the stakeholders ask the wrong ques-
tionships. According to certain theories of ontology, tions about the phenomena. On the other hand, the
however, some of these phenomena are misclassified; figure might be useful for database designers trained
for example, “landing rule” is not an entity but a law; to think in terms of the tables in a relational database.
“flight” is not an entity but an event; and “airport It would, however, have limited usefulness as a means
type” is not an entity but an attribute. of determining whether the focal stakeholders’ con-

88 October 2003/Vol. 46, No. 10 COMMUNICATIONS OF THE ACM

ception of a domain was captured faithfully.

Conclusion
Perhaps the best way IS professionals can facilitate
the validation of conceptual models is to generate
high-quality conceptual models from the outset.
Theories of ontology help ensure they select a con-
ceptual modeling grammar needed to produce high-
quality models of the focal domain. They can help
guide how the grammar is used to generate clear,
complete descriptions of the domain. And they can
be used to help make sense of ambiguous semantics
in conceptual models that need to be validated. c
PERHAPS THE BEST WAY
IS PROFESSIONALS
References CAN FACILITATE
1. Bodart, F., Patel, A., Sim, M., and Weber, R. Should optional proper-
ties be used in conceptual modeling? A theory and three empirical tests.
Info. Syst. Res. 12, 4 (Dec. 2001), 384–405. THE VALIDATION OF
2. Bunge, M. Treatise on Basic Philosophy, Vol. 3. Ontology I: The Furni-
ture of the World. D. Reidel Publishing Co., Boston, 1977. CONCEPTUAL MODELS IS
3. Burton-Jones, A. and Weber, R. Understanding relationships with attrib-
utes in entity-relationship diagrams. In Proceedings of the 20th Annual
International Conference on Information Systems (Charlotte, NC, Dec.
TO GENERATE HIGH-
13–15). Association of Information Systems, Atlanta, 1999, 214–228.
4. Chen, P. The entity-relationship model: Toward a unified view of data.
QUALITY CONCEPTUAL
ACM Trans. Database Syst. 1, 1 (Mar. 1976), 9–36.
5. Connolly, T. and Begg, C. Database Systems: A Practical Approach to MODELS FROM THE
Design, Implementation, and Management, 3rd Ed. Addison-Wesley,
Harlow, England, 2002. OUTSET.
6. Parsons, J. and Wand, Y. Emancipating instances from the tyranny of
classes in information modeling. ACM Trans. Database Syst. 25, 2 (June
2000), 228–268.
7. Parsons, J. An information model based on classification theory.
Manag. Sci. 42, 10 (Oct. 1996), 1437–1453.
8. Shanks, G., Nuredini, J., Tobin, D., Moody, D., and Weber, R. Rep-
resenting things and properties in conceptual modeling: An empirical
evaluation. In Proceedings of the 11th European Conference on Informa-
tion Systems (Naples, Italy, June 19–21). Universita di Napoli, Federico
II, Naples, 2003.
9. Simsion, G. and Witt, G. Data Modeling Essentials: Analysis, Design and
Innovation, 2nd Ed. Coriolis, Arizona, 2001.
10. Wand, Y., Storey, V., and Weber, R. An ontological analysis of the rela-
tionship construct in conceptual modeling. ACM Trans. Database Syst.
24, 4 (Dec. 1999), 494–528.
11. Wand, Y. and Weber, R. On the ontological expressiveness of informa-
tion systems analysis and design grammars. J. Info. Syst. 3, 4 (Oct.
1993), 217–237.

Graeme Shanks ([email protected]) is a

professor in the School of Business Systems at Monash University,
Clayton, Victoria, Australia.
Elizabeth Tansley ([email protected]) is a lecturer in the
School of Computing and Information Systems at Central Queensland
University, Rockhampton, Queensland, Australia.
Ron Weber ([email protected]) is a professor in the Faculty of
Business, Economics and Law at The University of Queensland,
Brisbane, Australia.

This article is based on research funded in part by an Australian Research Council

Discovery Grant.

Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. To copy otherwise, to republish, to post on servers or to redistribute
to lists, requires prior specific permission and/or a fee.

© 2003 ACM 0002-0782/03/1000 $5.00

COMMUNICATIONS OF THE ACM October 2003/Vol. 46, No. 10 89