AS & S review: OWL DL as RDF graphs

Section 4.2 Definition of OWL/DL Ontologies in RDF Graph Form
Version of 14 January
Review comments.

The first part of this section defines OWL DL ontologies 
in RDF graph form in terms of the outcome of the mapping 
from abstract syntax.
The remaining, largest part of this section is new.
It describes OWL DL purely in terms of RDF triples, independent
of the abstract syntax.
Although informative, this part is important, as it relates, 
for example, to the possibility to check whether certain RDF input
is OWL DL/Lite (Peter and Jeremy took actions about this in 
Manchester).

This section does not yet include everything that is required: 
- It does not yet contain similar definitions of OWL Lite 
ontologies in RDF graph form.
- It does not yet contain anything about AllDifferent.

As to the first part, I wonder about the condition labeled 4:
"the abstract syntax form provides a type for every individualID".
This condition is completely different from the other conditions.
It seems that it could be removed simply by filling in the
open hole in the transformation table of section 4.1 as follows:
S = individualID
T(S) = individualID rdf:type owl:Thing .
M(T(S)) = individualID
I would be in favor of this change: it would also make the
transformation to triples more symmetric with that of classID
and datatypeID.

An issue with respect to the second part is the correctness
of the presented pure-RDF description of OWL DL.
Another issue is that with the current ordering of paragraphs,
there are extremely many forward references, which makes
understanding difficult.  In particular, the definition of a
"description in G" makes use of most definitions that follow.
Moreover, several definitions, in particular that of description in G,
are recursive, which puts another burden on the plausibility
that this material is correct.
Yet another problem is that only during the treatment of the 
material it becomes gradually more clear what "meta scheme" is used:
there are definition triples and assertion triples.  I think I
may assume (this is not explicitly stated) that assertion triples 
include definition triples.

For the preceding section, 4.1 Translation to RDF graphs, it
is natural to follow a top-down approach, since the abstract
syntax is used as a starting point.
I believe that it is more natural to reorganize this section, 4.2,
following a bottom-up approach.
Then many parts could be understood to be correct, given
the translation from Section 4.1, before they are used in
the more "aggregate" parts.
Another argument to reshuffle the paragraphs of this section
is the final definition of OWL/DL graph, an acyclicity condition
appears: the definition triples of descriptions in G are required
to form a tree.  It is natural to let this definition be directly
preceded by the definition of descriptions in G.

Summary of these suggestions:
- reorder paragraphs so that no forward references appear
- provide description of "meta scheme" first: definition triples,
assertion triples etc.

I expect that this material could become more convincing in this
way.

--
It is confusing that in Section 4.2, unlike Section 2,
classes are excluded from descriptions and datatypes are 
excluded from data ranges.
Now the phrases "description or class" and "data range or
datatype" appear several times.
I believe that it is possible and desirable and that it could
slightly simplify the treatment to make this consistent with 
Section 2, where descriptions include classes and data ranges 
include datatypes.

Final comment, for now: the addition "in G" is done only in a 
few definitions.  It should be added to each definition, to 
make the distinction with the abstract syntax.


Herman ter Horst
Philips Research