From: AAAI-94 Proceedings. Copyright © 1994, AAAI (www.aaai.org). All rights reserved.
GICR A Genetic Model of Knowledge Representation
Angblica de Antonio, J&&isCamlefiosa, I&c Martinez Nonnand
Laboratorio de Inteligencia Artificial.
Facultad de Informirtica. Campus de Montegancedo.
28660 Boadilla de1 Monte. Madrid (Spain)
E-Mail:lia@fi.upm.es
Extended
Abstract
In the 1956 Darmouth College conference two aspects of
the definition of AI were emphasized: a) the separation
between the knowledge and the procedures using it and b)
the equivalence of the different knowledge representation
(KR) formalisms. Taking the last concept as an origin, an
idea arose in Knowledge Engineering:
Building generic
KR’s that could allow to represent any Knowledge Base
(KB) developed using any formalism, to work with it
without worrying about the actual formalism used in the
construction
of the KB. This is an objective that has not
yet been reached, although research in this area continues
as shown in the following examples:
- In the area of Validation and Verification (V&V) of
Knowledge
Based Systems (KBS) we can mention the
VALID project (ESPRIT II number 2148 project [CARD931). This project was based on the idea of building a
generic model of KR called CCR (Common Conceptual
Representation)
in which the formalism of any system
could be translated to apply a set of V&V tools to the
translated KB.
- In the Knowledge Acquisition area this idea has been
in the ACKNOWLEDGE
project
used, for example,
(ESPRIT II number 2576 project [ACK-881). The main
objective of this project was to develop a Knowledge
Engineering
Workbench
integrating
several knowledge
acquisition
methods, techniques
and tools. In order to
integrate the knowledge acquired by each of those, it was
necessary to use a generic KR called CKR (Core KR).
- Finally, this idea has also been used in Automatic
Translation. This idea is de basis of the INTERLINGUA
&presentation
(used in project PIVOT [NEC-861) which is
a representation
of the natural language knowledge independent of the actual language (Spanish, English, etc.)
used.
We show in this paper a proposal for a new generic
model of KR called GKR (Generic Knowledge Representation). This model has been developed as a result of the
analysis of the models described in the preceding examplcs. The study of the successes and shortcomings of these
models helped us to define GKR with several properties
that improve its representation
ability:
1438
Student Abstracts
- We have divided the representation of a KBS into three
parts: 1) a static part that represents the knowledge that has
the system about its problem domain (that is, the KB), 2)
a dvnamic part that represents, using traces of the execution, how does work the KBS faced to a problem (or test
case) and 3) some information referring to design particularities of the KBS. This part represents why does the static
part work as shown by the dynamic part. This part of the
systems represents control information.
- We have chosen frames [MINS-751 and rules [MAT&
881 as KR formalisms for the static part. These formalisms
are defined in GKR with characteristics
that were not
implemented in the other models, such as: representation
of user-defined facets, explicit representation of inheritance
rules, representation of non hierarchical relations between
frames and the representation
of conditions and actions
allowing rules to access or modify any part of the KB.
The above properties make possible to represent in GKR
things that would not be able to represent in the other
models. The definition of the GKR design is composed bJ
a set of structures that cannot be described in this abstract.
Although GKR can be used in other AI areas, it is being
used in the definition of a Validation environment based in
this representation
model. This environment
will apply
several V&V tools to KBS represented in GKR and it is
being developed by the Validation
Group of the AI
Laboratory of the Universidad PolitCcnica of Madrid.
Refemnces
[ACK-881 ACKnowledge Project. “ACKnowledge Technical Annex.” 1988.
[CARD-931 Cardefiosa, J. and Juristo, N. “General Overview of the Valid Project.” Proceedings of the European
Symposium on the Validation and Verification of KBS,
EUROVAV’93. Palma de Mallorca. Spain. 1993.
[MAT&881 Mate, J.L. and Pazos, J. “Ingenieria, de1
Conocimiento: disefio y construcci6n de sistemas expertos.”
Ed SEPA. 1988.
[MINS-751 Minsky, M.. “A framework for Representing
Knowledge” en “The Psychology of Computer Vision.” P.
H. Wilson (ed.) McGraw-Hill.
1975.
[NEC-861 Net. “Overview of Pivot”. C&C systems researchs laboratory. NEC Corporation. Japan. 1986.