Adapt OWL as a Modular Ontology Language
(A Position Paper)?
Jie Bao and Vasant Honavar
Arti¯cial Intelligence Research Laboratory,
Department of Computer Science
Iowa State University, Ames, IA 50011-1040, USA
fbaojie, honavarg@cs.iastate.edu
Abstract. OWL, the most commonly used web ontology language, has
limited provision for handling modular ontologies. Speci¯cally, the owl:imports
construct for linking ontology modules lacks support for partial reuse
of or localized semantics for the linked ontology modules. We propose
an alternative to owl:imports, namely, semantic importing, for linking
multiple modular ontologies with support for localized semantics, partial
ontology reuse, and distributed reasoning.
1 Introduction
The semantic web relies on shared ontologies for interoperability of a network
of loosely coupled, distributed, autonomous information sources and services.
Building large ontologies requires collaboration among multiple autonomous in-
dividuals or groups. Because ontologies designed with speci¯c domains, applica-
tions (e.g., medicinal chemistry), or users in mind often need signi¯cant changes
or additions before they can be successfully deployed in a related but di®erent
setting (e.g., translational medicine), there is an urgent need for support for
selective reuse of independently developed ontology modules.
OWL1 provides the owl:imports construct for linking multiple OWL on-
tologies to form a larger OWL ontology. However, such a syntactic importing
solution of OWL su®ers from several limitations:
{ Lack of support for localized semantics. As Bouquet et.al. [5] have previously
observed, OWL provides only a global semantics for an ontology module that
satis¯es all the axioms and facts in all ontology modules that are(directly
and indirectly) imported by that module. Thus, the importing of a module
B by module A is tantamount to the copying of all statements of module B
into module A.
{ Lack of support for partial reuse of ontology module. The lack of support
for selective reuse of parts of one ontology module limits the ability to reuse
ontology modules because using the module in its entirety might introduce
unwanted inconsistencies or impact performance (because of the presence of
a large number of assertions that might be irrelevant in the context of the
target application).
? This research was supported in part by the US NSF award 0639230
1 http://www.w3.org/TR/2004/REC-owl-ref-20040210/

2Consequently, there is a growing interest in OWL language constructs (or
language extensions) to support modular ontologies, including several syntactic
extensions to OWL including dOWL [1], C-OWL [5] and E-Connections [6].
However, they are also limited in several ways:
{ Limited expressivity [2]. C-OWL does not support linking two classes in
di®erent modules with roles, and E-Connections has no direct support for
inter-module class subsumption.
{ Reasoning di±culties [2] arising from the lack of mechanisms to prevent
arbitrary domain relations in C-OWL and the requirement of strict domain
disjointedness in E-Connections.
Against this background, we argue for a semantic extension to OWL, i.e.
a new interpretation of owl:imports grounded in modular semantics which
allows the connected ontology modules to have partially overlapping local inter-
pretations to ovecome some of the existing limitations of existing approaches to
linking OWL ontology modules.
2 A Modular Semantics for owl:imports
Instead of introducing syntactic extensions to OWL, we propose a new se-
mantics for owl:imports based on an interpretation of owl:imports that pro-
vides direct support for modular ontologies. Our proposal di®ers from existing
approaches in several respects:
1. support for both inter-module class subsumptions (e.g. i : C v j : D) and
inter-module role relations (e.g. i : C v 9i(: r):(j : D)); use of importing
approach instead of the linking approach used in C-OWL and E-Connections;
2. relaxation of the strict local domain disjointness that is required by E-
Connections (thereby avoiding the associated reasoning di±culties arising
from such an assumption).
The result is an approach that o®ers a modular semantics for OWL ontologies
that instead of requiring the ontology modules to share a completely overlapping
model, allows them to have only partially overlapping models. The basic intuition
is that when a symbol is shared by di®erent modules, the restrictions on the
interpretation of that symbol must also be shared by those modules.
Formally, given a set of OWL ontology modules fOigi2I , we say a module
Oj imports a term i : t if Oj (directly or indirectly) imports Oi and refers i : t
(denoted as i t¡! j). A modular semantics for OWL can be speci¯ed as:
De¯nition 1 (Modular OWL Interpretation) . A modular OWL interpre-
tation for a set of ontology modules fOigi2I is a family I = fIigi2I , where each
Ii = h¢i; (:)Iii is the local interpretation of Oi. For any term importing relation
i t¡! j, we have tIi = tIj .

3The resulting semantics di®ers from the current OWL semantics in that it
does not require the complete overlapping of local domains, i.e., ¢i = ¢j , for
any i; j. It also relaxes the (explicit or de facto) domain disjointness adopted
by E-Connections and C-OWL. Thus, it o®ers a selective importing mechanism
that allows the parts of an imported ontology module that are selected for reuse
by another module to share their interpretation with that module whereas the
other parts of the ontology (i.e., those that are not selected for reuse) retain
their local interpretations.
Example: Consider two ontology modules modelling domain knowledge about
wine and food, respectively. Suppose the food module contains the following
terms and axioms:
food : Apple v food : Fruit (1)
food : Grape v food : Fruit (2)
Suppose the wine module imports the food module and contains axioms:
wine : WineGrape v food : Grape (3)
wine : Wine v 9wine : madeFrom:(food : Grape) (4)
An interpretation of the ontology contains two local interpretations:
{ I1: food : FruitI1 = fx1; x2g; food : AppleI1 = fx1g; food : GrapeI1 = fx2g
{ I2: food : GrapeI2 = wine : WineGrapeI2 = fx2g; wine : WineI2 = fx3g;
wine : madeFromI2 = fhx3; x2ig
Hence, the importing relation from food to wine is not complete, but partial
in that only the terms selected for reuse (e.g. food:Grape) are interpreted in the
shared part of local domains (fx2g in the example), whereas the terms that are
not selected for reuse (e.g. food:Apple) retain their local interpretations.
The example also shows that the semantic importing strategy supports both
inter-module class subsumptions (e.g. axiom 3) and inter-module role relations
(e.g. axiom 4).
3 Reasoning with Modularized owl:imports Semantics
Package-based Description Logics (P-DL) [2] adopts the semantic importing
approach that allows local domains of ontology modules to be partially overlap-
ping. A distributed reasoning algorithm for P-DL ALCPC , i.e. ALC extended
with concept importing between ontology modules has been described in [3].
Recently, this approach to reasoning with modular ontologies has been shown
to work with modular counterparts (e.g., SHOIQPC) of fairly expressive de-
scription logics SHOIQ wherein SHOIQ (OWL-DL) modules are connected
using concept name importing [4]. Hence, if we were to adopt OWL with the
proposed new owl:imports semantics as the syntax for P-DL, or equivalently,
adopt the P-DL semantics for OWL-DL, we can o®er support for selective reuse
of modular ontologies, which ensures that individuals that are imported from one
module into another module (through the owl:imports mechanism) to share a
common interpretation. It thereby ensures that the inferences resulting from a

4distributed reasoning process are identical to those that are obtained from an
integrated ontology that combines the selected parts of the relevant ontologies
into a single centralized ontology [2] .
Pan et.al. [7] have recently proposed an extension to OWL that allows both
syntactic importing (using the owl:imports construct) and semantic importing
(using a new owlx:semanticImports construct). However, reasoning in such a
setting, unlike in the case of our proposal [3], is limited to a pair of ontology
modules with no cyclic importing.
4 Conclusions
The proposed modular semantics for owl:imports o®ers a promising ap-
proach to supporting partial (and selective, and hence context or application
speci¯c) reuse of existing ontology modules. In particular,
{ It supports both inter-module class subsumption and inter-module property
relations.
{ Because it requires no new syntactic extensions to OWL, it ensures backward
compatibility of the resulting modular OWL ontologies processable using
existing tools in settings that do not require support for selective reuse of
ontology modules).
{ It allows parts of an ontology module that are not shared with another mod-
ule to retain their local interpretations thereby o®ering a practical tradeo®
between complete domain overlap and strict domain disjointness.
{ It supports sound distributed reasoning among a federation of loosely cou-
pled ontologies.
Work in progress is aimed at tool support for OWL-based modular ontolo-
gies, such as an ontology ontology editor that supports collaborative building
of modular OWL ontologies, and a reasoner for distributed reasoning under the
proposed OWL semantics.
References
1. Avery, J., Yearwood, J.: Dowl: A dynamic ontology language. In: ICWI. (2003)
985{988
2. Bao, J., Caragea, D., Honavar, V.: On the semantics of linking and importing in
modular ontologies. In: I. Cruz et al. (Eds.): ISWC 2006, LNCS 4273. (2006) 72{86
3. Bao, J., Caragea, D., Honavar, V.: A tableau-based federated reasoning algorithm
for modular ontologies. In: Accepted by 2006 IEEE/WIC/ACM International Con-
ference on Web Intelligence (In Press). (2006)
4. Bao, J., Honavar, V.: Reasoning with expressive modular ontology languages. Tech-
nical report, Computer Sicence, Iowa State University (2006)
5. Bouquet, P., Giunchiglia, F., van Harmelen, F., Sera¯ni, L., Stuckenschmidt, H.: C-
owl: Contextualizing ontologies. In: International Semantic Web Conference. (2003)
164{179
6. Grau, B.C., Parsia, B., Sirin, E.: Working with multiple ontologies on the semantic
web. In: International Semantic Web Conference. (2004) 620{634
7. Pan, J., Sera¯ni, L.: Semantic import: An approach for partial ontology reuse. In:
WoMo 2006, co-located with ISWC. (2006)