Modeling and Information Processing
MC2 - Modeling and capitalizing on knowledge
Formal ontology and interoperability by semantics
By ontology, we mean "Common semantics in a field shared by a community of interest" and by "Formal ontology" we mean a computational and computerized ontology on the basis of which processing and transformations can be done. The interoperability of information systems poses two difficulties, amongst others:
- How can we use a formal ontology to resolve the exchange mechanisms between heterogeneous processes that must share data and knowledge but whose semantics are partially different?
- How can we ensure the coherence and consistency of the different results produced by these heterogeneous processes?
We present two pieces of work in which we used compilation techniques to provide an answer.
The Oasis system
This system is one of the results of the DEFR [1] project, whose area of application is the fusion of intelligence. Oasis is an experimental, generic and extendable platform for the fusion of multi-source intelligence. The data processed by Oasis has a high semantic level. The fusion process is designed as a heterogeneous algebra of fusion operators (classification, correlation, inference by fuzzy logic or temporal logic, etc.) Each fusion operator processes abstract data types. Oasis uses three theories: descriptive logic, fuzzy logic and pattern matching associated with database searches by SQL queries. These operators are generic because they are independent of the semantics carried by the data and depend only on the data types handled. The semantics of the data and knowledge is explicitly borne by the input ontology. This includes descriptive models of concepts. The ontology of Oasis includes battle commands (models and instances) and the a priori entities of the battlefield or the crisis (units, equipment, activities, etc.).
To formalize the ontology, we have designed what we call Roc language that the user uses to describe entities. Roc captures and describes the data types necessary for each fusion operation. We use compilation techniques to build a compiler that takes the descriptions in Roc language as inputs and generates:
- different models necessary for each of the inference engines corresponding to the three theories mentioned above,
- database schemes (DBMS O2) supporting the ontology,
- html files for the end user documentation.
This approach allows us to unify the fusion algorithms. Interoperability is a consequence of this ontology-based approach. Coherence is ensured by the automatic generation from a single textual source file containing the concepts and entities. The consistency of the combinations of the results produced by the fusion operators is made possible by common data semantics and shared reference concepts, and not the shared data in the database
An example of a description in the ROC language
The Genesis tool
The second example concerns simulation for Extended Air Defense systems. The project's objective is to define and write the Genesis software that can be used to build EAD simulation federations very rapidly on a network of PCs or Unix stations via the HLA communications protocol. The simulators or federates, in HLA terms, are considered to be black boxes, only known by the data exchanged: Objects (sort of C++ object) and interactions (named set of parameter values). The federation is made up of interoperable simulators, or federates, that may be predefined or adapted or built. To do this, we have defined the federate description language, called LD, used to specify:
- the date exchanged: objects, interactions and exchange methods: static / periodic / conditional, pull / push, etc.
- the federates: import of objects and interactions, synchronization points, events, scenario, etc.
- the federation: the set of federates executing on a network.
HLA concepts, objects, interactions, federates, etc. make a "sort of" type hierarchy that constitutes an ontology supporting the semantics shared by the federates. Genesis operates in two stages:
- It analyzes the files in LD and collects the data. It then detects any errors in syntax and semantics.
- t generates all or part of the different codes necessary for producing the federation: SOM and FOM for the designers, the HTML documentation for the end user, the different federates' C++ files and compiling and link editing makefiles, the federations' configuration and start of execution files.
The development method's advantages are the following:
- The coherence and consistency of the ontological concepts are integrated into the compiling process. Semantic tests are automatically produced based on the ontology of the types, concepts and entities and make up a database of tests for maintenance.
- Total control of software development. The cost of integrating new functionalities can be evaluated. These are easy if it is a matter of adding new concepts or new instances of concepts without adding a root ontological concept, if not you must intervene at language level to take the new ontological roots into account. In both cases the development process is still controlled.
The interoperability is a consequence of the use of an ontology associated with compiling techniques and automatic code generation.
References
[1] Véronique Royer - ONERA/DTIM Châtillon, FR, Jean-François Challine - Thales – TCC/IDS/DTS Colombes, FR, A platform for interoperable fusion model, IJCAI 2001
[2] Jean Bourrely - ONERA/DPRS, Michel Barat - ONERA/DTIM, Manuel Utilisateur Genesis, DGA/STTC N° 02.34.009/MU/001-1.10
