KNOWLEDGE SHARING 

Position Statement         Harold Boley, DFKI         Ellivuori, 26 May 1998



Knowledge sharing involves the porting of knowledge(-module) libraries
across user platforms. A knowledge module may contain object and method
declarations, logical facts and rules, or natural-language statements.

High-quality conceptual modelling is desirable for many reasons, but
is inevitable for structured knowledge sharing. Like metadata, such
models act as hierarchically organized catalogs (`ontologies') that
support knowledge sharing (access, evolution, integration):

* Obtaining and keeping an overview of a knowledge library (access)

* Finding knowledge modules/items via semantic descriptions (access)

* Incrementally updating knowledge items (evolution)

* Consistently merging knowledge modules (integration)


Here is a (still incomplete) list of twelve of the current research issues:

1 Top-level ontologies and ontology alignment: structuring knowledge sharing?

2 Granularity of sharing: Many small modules or a few big modules?

3 De- and recontextualization of knowledge modules/items: fore- vs. background?
 
4 Descriptive terms: global (e.g. RDF) or local (enterprise) standard?

5 Formalism for semantic description: attributes (OOP), relations (LP), ...?

6 Description formalism and content language: same or different notation?

7 Intersection/union languages: light- (e.g. Datalog) / heavyweight (e.g. KIF)?

8 Expressiveness hierarchy: conformance levels (e.g. Relfun) or profiles (KIF)?

9 Separate knowledge kinds (e.g. inference rules vs. integrity constraints)?

10 Translation between all, or from/to canonical languages (e.g. KIF, CGs)?

11 Interchange without integration: wrapper (e.g. KQML), not content language?

12 Program/knowledge reuse/sharing: which lessons learned can be transferred?


A number of American and international committees are working on standards
for knowledge sharing. Conceptual modelling and ontologies play a central
role in these efforts.

Too little work is being done on (empirically evaluating) real knowledge-
sharing projects. Definitive answers to most of the twelve questions can
only be given afterwards. So, why don't we start with some international
knowledge-module library of shared interest in the Web?