Re: UFDTF Metamodeling Document

The RDF metamodel in particular was reviewed extensively by HP Labs, Pat 
Hayes, numerous IBM Watson folks, other OMGers who are metamodeling 
experts or have implemented RDF-based applications, and while I will go 
back to them with your comments, below, to verify the metamodels and 
related text again, I believe the RDF metamodel chapter of the ODM 
specification is correct with respect to most, if not all of your 
comments, below.

For example, graphs can span multiple documents, multiple graphs can 
occur in the same document, the relationship between statements 
(triples) and graphs in particular was reviewed extensively and was the 
subject of significant debate for several weeks.  The RDF documents are 
unclear about a number of the issues you've raised, thus we worked with 
the RDF community, and in particular HP Labs, to ensure that the ODM was 
correct and as unambiguous as possible.  The same is true for the 
reification language and approach, use of the word statement rather than 
triple (which we also debated at length, and which Pat and Chris Welty 
assisted us with may be a matter of taste -- it is used interchangeably 
in many places in the RDF documents), multiplicities (which may appear 
to be unintuitive to someone less familiar with MOF as you mention), 
optional naming of graphs (per SPARQL), and so forth.

One option with regard to OWL would be to provide extensions to the 
current OWL metamodel for those who depend on RDF, corrections where 
required, and, maybe -- I can't commit to this without discussion in the 
OMG community -- an additional OWL DL only metamodel, divorced from the 
RDF metamodel, for those, such as some members of this working group, 
who need the standalone approach for some of the applications that Boris 
named.  This could be done in a revision to ODM, once the language 
developed by this community is sufficiently stable, and in a way that 
would clearly provide conformance requirements to users, the 
relationship between the two metamodels, and so forth.  Having said 
this, I stand by my comment that we need to be thoughtful about what we 
would do here to limit confusion among those in the community who have 
already implemented, are in the process of implementing, or may 
implement tools that use the specifications.  This community is, in 
fact, significantly larger than you might imagine -- which we learned 
only from statistics on downloads of presentations, etc.  One 
presentation given early this year in particular was downloaded over 
2000 times within a few short weeks.

Thanks,

Elisa

Peter F. Patel-Schneider wrote:

>Unforunately, I think that there often cannot be a distinction between
>comments on the metamodels and comments on the related text, as the
>related text serves to illustrate and refine the situation expressed in
>the metamodels. 
>
>For example, consider "URIREferenceNode, BlankNode, and RDFLiteral form
>a complete covering of RDFSResource and are pairwise disjoint."  This
>contains one very wrong part and one very right part.  If the right part
>(pairwise disjointness) is ignored, then the metamodel is incorrect.
>The very wrong part is that RDFSResource is a syntactic superclass of
>URIREferenceNode, BlankNode, and RDFLiteral, but this is also reflected
>in the metamodel itself.   
>
>The basic problem is that there is a fundamental misconception of the
>relationship between RDF syntax and semantics embedded in the RDF
>metamodel.  Fixing this misconception appears to me to require a
>complete rewrite of the metamodel and accompanying text.
>
>Here are some other problems in the metamodel:
>
>- the metamodel use statements (instead of triples)
>- the metamodel does not say anything about the uniquess of statements (triples)
>- the metamodel allows statements (triples) to be in multiple graphs,
>- the metamodel has a unique reification status for each statement (triple)
>
>As well, the metamodel has the wrong cardinalities for the relationships
>between resources (nodes) and comments/labels and between typed literals
>and datatypes.  (I'm not prepared to say exactly what is wrong, because
>I seem to remember that UML diagrams have the cardinality restrictions
>on the other end from that which would be intuitive.)  In actuality a
>node can have any number of comments/labels in RDFS and any string can
>be used for any number of comments/labels.
>
>peter
>
>
>From: "Elisa F. Kendall" <ekendall@sandsoft.com>
>Subject: Re: UFDTF Metamodeling Document
>Date: Thu, 29 Nov 2007 12:56:23 -0800
>
>  
>
>>Just to follow up on Conrad's response, I would really encourage you to 
>>submit comments to the issues list, which would help us improve the 
>>specification in the short time remaining.
>>
>>I think some of the issues you raised are legitimate, I disagree with 
>>others, and some comments may stem from lack of knowledge of MOF, 
>>particularly its purpose as a repository/persistence representation. 
>>    
>>
>
>Which are which?
>
>  
>
>>Most are not comments on the metamodels themselves but on the related 
>>text, which can be fixed where appropriate. Metamodel changes are 
>>harder, due to the stage in the standardization process and number of 
>>implementations, but possible, depending on their nature. We should 
>>simply work through them, in the OMG forum rather than on this list, and 
>>rather than requiring another three years to develop an entirely new set 
>>of models.
>>    
>>
>
>Well, I am afraid that I am of the opinion that it would be better to
>start over.  I agree that it may be better to has this out in some other
>forum.  However, I don't have management support for me doing
>OMG-related work.
>
>  
>
>>I would also like to say that the use cases we've been discussing in the 
>>user facing documents TF are essential to teasing out metamodel 
>>requirements, 
>>    
>>
>
>How so?
>
>  
>
>>which in turn, should have beneficial input to the 
>>language development.  We worked synergistically with the Common Logic 
>>language developers in creating the metamodel for CL, which had clear 
>>impact on the language itself.  The CL metamodel in the ODM 
>>specification is the metamodel in the ISO CL specification. Any approach 
>>that provides a smorgasbord of diagrams -- UML conceptual view in the 
>>structural specification, a different metamodel in a W3C note, and a 
>>third international standard metamodel in the ODM, will only serve to 
>>confuse and divide the user community. 
>>    
>>
>
>Well, maybe, but it may also be that the three set of "diagrams" serve
>entirely different purposes and thus deserve to be different.  It may be
>that the OMG OWL diagrams are designed to support light-weight
>OWLish processing of RDF graphs instead of actually supporting OWL, and
>thus my comments on them not being appropriate for OWL are not of
>interest to the OMG.
>
>  
>
>>Elisa
>>    
>>
>
>
>peter
>
>
>
>
>  
>

Received on Friday, 30 November 2007 13:37:43 UTC