cycles in subclass hierarchy (was: RDFCore Update)

Brian McBride <> wrote:

> The minutes of the RDFCore WG face to face meeting
> have just been approved 

The following is a summary of the discussion in the DAML+OIL joint
committee as input to the decision on allowing cycles in the RDFS 
class hierarchy or not.

The message is: allowing cycles is mandatory for DAML+OIL, and not
allowing them in RDFS will seriously break the layering of the
Semantic Web architecture.

In more detail:

1. in DAML+OIL, subclass-relations can be inferred even if they are not
explicitly stated (note that this is an important difference from RDF-S,
where A is only a subclass of B if >*and only if*< it is either
explicitly stated, or follows from subsumption in the class-hierarchy).

2. as a result of point 1, acyclicity of the subclass-relation in
DAML+OIL cannot always be detected by purely syntactic means, since
computationally expensive inference may be needed to detect implicit
("implied") subclass-relations
3. because of point 2, DAML+OIL cannot require acyclicity of the
subclass-relationship, since that requirement could only be enforced at
great computational expense.
4. Point 3 means that if RDF-S will enforce acyclicity of rdfs:subClassOf, 
then DAML+OIL can no longer use rdfs:subClassOf.
In other words, DAML+OIL will be forced to introduce daml:subClassOf

5. Point 4 would mean that much (all?) backward compatability between
RDF-S and DAML+OIL would disappear: an RDF-S processor would be unable
to catch any of the semantics of a DAML+OIL ontology (whereas currently,
all the explicitly stated subclass relationships in a DAML+OIL ontology
are accessible to an RDF-S agent, since DAML+OIL uses rdfs:subClassof

6. An important design rationale behind DAML+OIL (and in fact much other
stuff on the Semantic Web so far) has been a layered approach, where
languages are stacked on top of each other, with as much partial
interpretation between the layers as possible. Tim BL has even argued in
[1], [2], [3] and many other places that such "partial understanding is
an essential design principle of the Web in general, Semantic or not.

The decision to make rdfs:subClassOf acyclic will force DAML+OIL to
introduce daml:subClassOf, and will therefore lead to an almost total
loss of partial understanding between these two closely related ontology

Frank van Harmelen.
Department of AI, Faculty of Sciences,  Vrije Universiteit Amsterdam
de Boelelaan 1081a, 1081HV Amsterdam, The Netherlands
tel (+31)-20-444 7731 fax&voicemail (+31)-84-8722806

[1] Evovability, WWW7 keynote speach, 
[2] "Web Architecture from 50,000 feet"
[3] "Web Architecture: Extensible languages"

Received on Monday, 3 September 2001 16:59:49 UTC