[RDFTM] Comments on the recent editor's draft from the ODM working group

Steve and all,

Because the issues that RDFTM is attempting to address are also relevant 
to our work on ODM,
we felt that it was important for us to review and comment on this 
document.  As you and Lars
may recall, Lewis Hart, formerly with AT&T Government Solutions, did 
much of the original work
on our metamodel for Topic Maps.  More recently, though, Bob Colomb has 
taken on much of this
work, with significant input from the ISO TM community, particularly in 
Japan.

We discussed the issues during our conference call this week, summarized 
as follows.

1. The OMG ODM (Ontology Development Metamodel) group has been working 
on mapping among various ontology representation languages for nearly 
two years. This work includes mapping between Topic Maps and OWL, so the 
RDFTM proposals impact the ODM.

2. Even though the RFP for the ODM called for normative mappings, we 
have concluded that normative mappings are not practicable, for reasons 
detailed in Appendix E of the November 2005 draft submission [1]. These 
reasons include the subtle incompatibilities among languages which are 
the subject of the recent exchange with Peter Patel-Schneider [2]. The 
ODM team therefore argues that whatever the eventual result of the RDFTM 
working group, that the product be seen as informative, rather than 
normative.

3. The ODM team is happy with the concept of a semantic mapping. The ODM 
mappings are also semantic. However, we take issue with the choice of 
guided rather than unguided mappings. Of course unguided mappings are 
underdetermined. That is one of the reasons for the decision that the 
ODM mappings be informative only. We agree that in practice a mapping 
will need to be guided. Our objection is that the context-free guidance 
proposed for the RDFTM will likely have limited utility. Reasons for 
this objection are detailed below.

3.1 Using terminology from our usage scenarios analysis in Chapter 7 
(table 7) of [1], there are at least two reasons why one might want to 
map an ontology from one representation to another. One is that an 
ontology in one representation describes a legacy system, and that the 
original ontology, part, or all of the legacy system is being migrated 
to another representation. This would be typical of information systems 
development applications.  Here, the context for guidance is dependent 
on the aims of the specific project, which could easily differ from the 
generic choices included in the RDFTM.  Further, under such 
circumstances it would be unlikely that the continuing development would 
wish to maintain backwards compatibility with the legacy source.  This 
is partly because the subtle differences in expressibility make this 
difficult, and partly a cost measure.  So for this kind of application, 
the RDFTM guidance is very likely to be too generic, and the 
recommendation for annotation of the target unlikely to be maintained in 
downstream development.

3.2 Another use case considered requirements for continuing 
interoperability in run-time interoperation applications like e-commerce 
exchanges and application development applications involving re-use of a 
standard domain ontology like Gruber's Engineering Mathematics ontology 
or the Foundational Model of Anatomy. Here, the context for any mappings 
is the specific ontology governing the application. There are many 
different methods of representing ontologies, so it is possible that the 
governing ontology would be represented in a language other than RDF or 
TM. In particular, neither RDF nor TM are used for representing either 
the Engineering Mathematics or Foundational Model of Anatomy ontologies. 
The former was developed in Ontolingua and the latter in a frame-based 
language. So a project to develop an application in this space in either 
RDF or TM would have to make compromises and design decisions different 
from the guidance proposed in the RDFTM.

3.3 The ODM team therefore argues that the RDFTM describe the unguided 
mappings with their indeterminacies, then provide some guidance using a 
non-normative light-weight mechanism.

4. From the perspective of OWL, the annotation method advocated in RDFTM 
for RDF is the use of properties whose domain is other properties. The 
annotated ontology is therefore OWL Full. As noted in the RDFTM, OWL has 
a lighter-weight mechanism compatible with OWL DL, namely annotation 
properties. Although these properties cannot be used for property 
axioms, they can be used by the software implementing particular mapping 
languages.

5. The RDFTM includes a consideration of a number of untyped topic map 
constructs: associations, occurrences, association roles. The working 
party should know that in TMDM 2005-12-16 (and in the earlier 
2005-10-28) all these constructs are required to be typed.

6. Note that although RDF does not have a container construct like Topic 
Map, OWL does.  An OWL ontology, as distinct from the resource 
owl:Ontology, contains the statements defining it, providing context for 
the restrictions in OWL DL and OWL Lite.

Thanks and best regards,

Elisa

[1] http://www.omg.org/cgi-bin/doc?ad/05-09-08
[2]http://lists.w3.org/Archives/Public/public-swbp-wg/2006Feb/0099.html