Unifying MDA and Knowledge Representation Technologies

1
Unifying MDA and Knowledge Representation
Technologies

• Colin Atkinson
• University of Mannheim
• 20th September 2004

1st International Workshop on the Model Driven
Semantic Web
2
Motivation for Integration

• model driven development and knowledge
  representation are naturally converging
• software applications are increasingly expected
  to utilize domain knowledge
• enterprise systems are increasingly integrated
  from third party components (e.g. Web Services)
• web applications are becoming increasingly
  complex and thus in need of software engineering
• hence the growing number of integration
  initiatives
• OMG RFP for Ontologogy Definition Metamodel /
  Profile
• several papers
• this workshop .
• BUT How?

3
Existing Integration Initiatives

• Accepted wisdom
• MDA and ontology description technologies are
  inherently distinct technology spaces with at
  least some fundamentally distinct concepts and
  mechanisms
• the two technologies should be integrated using
  official OMG extension mechanisms to define an
  ontology representation language within the MDA
  framework
• Extension based approaches extend the MDA
  framework by either -
• defining a complete new metamodel (or profile)
• directly extending the UML metamodel
• indirectly extending the UML metamodel by means
  of profile

4
Isnt Ontology Definition a form of Modeling?

• what is the true relationship between modelling
  and ontology representation languages

• are the existing differences fundamental or just
  a quirk of personal taste and/or history?
• like the difference programming languages

5
The Properties Problem

• the usually quoted shortcoming of the UML for the
  purpose of ontology representation -
• it is not possible to show that different
  incarnations of a relationship (e.g. owns) are
  somehow the same

owns
owns
Landlord
House
Company
Car

• but although association inheritance is not
  directly supported, the effect can easily be
  programmed up

6
The Case for Core-Level Unification

• there is no fundamental difference between
  ontology representation and modelling
• the former is a proper subset of the latter
• the UML is already an adequate ORL
• any differences are historical and superficial
• defining an ODM implies that UML is not for
  ontologies
• adds to the confusion already felt by developers
• am I dealing with a model, data or knowledge
• core level unification greatly simplifies the
  message
• developers just represent information
• strengthens the formal foundation of the UML and
  the overall MDA framework (formal semantics)

7
What does Core-Level Unification Mean?

• rearchitecting the underlying MDA framework to
  better support ontologies
• enhancement of the infrastructure library core

• moving to a clean multi-level modelling framework

8
MDA Four Layer Model
9
Two Dimensional Modeling Infrastructure
Core Language
Object
Classifer

• rules

• "knowledge"
• - patterns

Structural Obj.
Class
Dog
instanceOf
instanceOf
Breed
Poodle
Fido
Poodle
Breed
Fido
10
Different Views of the Infrastructure
Tool Developer's View
11
The Role of ORL Profiles

• even if unification takes place at the core level
  there will always be a role for ORL profiles
• they will be needed to create platform-specific
  ontology representation
• the core modeling features will be used to
  create platform independent ontology
  representations
• ORL extensions are needed to support platform
  specific ontology representations
• e.g. Java profile, C profile, EJB profile ..
• the ORL is the platform

12
Conclusion

• there is no fundamental difference between
  modeling, ontology representation and data
  representation
• there should be just one core language for all
  three
• the unified infrastructure should be founded on
  multi-level modelling
• two fundamental modelling dimensions
• unification of classes and objects (clabjects)
• deep instantiation
• .
• ORL profiles are needed to support platform (i.e.
  ORL specific) ontology representations