6th European Systems Science Congress (Sept. 19-22, 2005)
******

Systems Science : the Ghost in the Ontologies ?

Keywords :

Conceptual modeling, Knowledge representation, Ontology, Terminological formalism, Semantic model

Extended Abstract; voir un bref résumé en français ci-dessous

1. Introduction

Semantic models came to be used to deal with problems ranging from database design to retrieval and interchange of an exploding amount of on-line information, over different goals of Artificial Intelligence where the researchers supported the same primitives and mechanisms.

Subsequently, ontologies are given as a consensual description for knowledge sharing. Escaping the philosophical foundations, computer science turns ontologies in a vast program of description of the concepts of particular domains as well as the relationships in which they are involved. However, the same entity may be defined in different ways while a common vocabulary may hide data heterogeneity.

Our interest in conceptual maintenance and ontologies alignment emphasizes, on the one hand, the intuitive aspect of the epistemological primitives of description and, on the other hand, the impact of the systemic features. It stands to reason since only cognitive science and systems science can give an ontological dimension to the data structures.

2. Methodology

In the paper we aim at critically inspecting semantic models, conceptual schemes (application–specific use of the previous ones) and their relationships to clearly circumscribe the underlying heterogeneity.

Our strategy consists in superimposing a syntactic construction. This addition reveals the "systemic ghost" (clarifies what is already present) and offers a way to add more systemic information.

After an overview of historic and panoramic foundations in the field of conceptualizations, our abstract structures are drawn closer to existing models in an attempt to study the implicit systemic features.

3. A quest to capture the meaning of data

3.1 Hierarchy and inverted-trees structure

The word "ontology" stands for semantic models with different degrees of structure ranging from simple hierarchies to elaborated knowledge representations. Thus far, the word "hierarchy" may be confusing.

The considered hierarchies are often taxonomies and a lot of work has been done concerning the is-a link (generalization). However, the part-of links may also create a hierarchy. Unfortunately, when examined in some detail, a systemic perspective points out different kinds of the part-of link. Moreover, our abstract construction may exhibit another type of hierarchy when the relation of inclusion is considered among different partitions. This step emphasizes the miscellaneous semantic interpretations needed for the inverted-tree structures.

3.2 Object characteristics and aggregation of attributes

Semantic models use attributes to construct or interrelate objects. As previously declared, our attribute is not user-specified to correspond to given characteristics but instead defined with a mathematical constructor.

During this step of abstraction, the distinction between subsets and systems vanishes. A fundamental consequence is to show that the aggregation of the attributes as used in the Codd relational approach is not orthogonal to generalization, the n-ary notation remains abstract and the part-of link may be suspect from a systemic point of view.

It is argued that when the constructed attributes receive a systemic interpretation, it predicts the influence of structure reformatting.

 3.3 Relationships, logic and holes

Many expressive notations have been developed, crossroads of networks and frames, they provide an object–centered associational representation with a natural graphical form. Since they are a subset of First Order Logic, they cannot be asked for representing more. The crucial worry is to know what is represented.

The paper provides a self-contained introduction to the notion of species of structure (defined in Bourbaki). It introduces explicit typification to point out that components and links are missing in such ontological engineering. The holes explain the computational intractability of knowledge transformations.

The relationships between the things are often described in a similar way independent of their nature. Moreover, the progressive elaboration of ontology with an associational methodology turns the intrinsic characteristics of the entities into links and the extrinsic characteristics surprisingly form the attributes.

4. Conclusion

The paper clarifies the implicit role systems science plays during the semantic achievements in a formal language of structure. During our process of clarification some new qualitative concepts are suggested to preserve the global comprehension of systems, often lost during the analytical findings. The resulting benefit stands as much for systemic modeling as for construction of ontologies.

5. References

According to the available room in the definitive paper, the bibliography will list seminal and foundational sources of knowledge representation with a higher priority ; the huge amount of applications found in books and international journals or conferences may be sampled for illustration purposes.

6. Annexes

The oral presentation will use examples, the definitive paper may accommodate them in such a section.

Short French summary

 Le but n'est pas ici d'entamer un parcours historique et philosophique mais un exposé technique tentant de répondre, grâce au regroupement de réalisations pratiques et de formalisations théoriques, à la question : "quelle part de représentation d'un système peut-on trouver dans une ontologie ?".

La construction des ontologies en informatique a clairement réduit les fondements philosophiques à une revendication de légitimité pour déboucher sur une vaste entreprise pragmatique. Les définitions et les pratiques (concernant les ontologies) qui en sont issues mettent en valeur un paradoxe :

Il est rare de revendiquer une pure logique de description (hiérarchie créée par différentiation), l'aspect relationnel est prôné comme "raison d'être" de l'ontologie. Il est tout aussi rare que cet aspect relationnel bénéficie d'une axiomatique voire même de fondements épistémologiques.

Une sorte de péché originel semble contraindre à choisir entre taxonomie et bric-à-brac. Notre thèse est qu'il s'agit d'une difficulté à expliciter [1] les systèmes dans les langages offerts par l'informatique. Si on l'admet, on comprend mieux les difficultés de la recherche en représentation des connaissances et le partage obligé avec les modèles subsymboliques reposant sur l'émergence et la non-explicitation.

Pour formaliser cette "difficulté à expliciter" nous allons successivement mettre en valeur l'absence ou la présence de "l'esprit systémique" dans différents aspects de la représentation des connaissances structurelles.

On confrontera ainsi l'arborescence avec la notion de hiérarchie. Cela conduira à poser la question de savoir ce qu'est une propriété et à rapprocher sa construction théorique des différents usages qui en sont faits. On en déduira le rapport ambigu que le standard de fait des bases de données dites "relationnelles" entretient avec la systémique. Les modèles d'inspiration "associationistes" (que l'on retrouve dans Merise ou UML, par exemple) seront remis à plat et leurs outils de description seront mis en perspective avec ceux des systèmes. On en déduira la forme que devrait prendre l'explicitation systémique mais aussi des outils pour la construction des ontologies.