Re: asn06/OWL vs. UML [was: asn06 take 2 (Abstract Syntax as a kind of ontology?)]

Gerd Wagner wrote:
>> I agree UML is genuine option.
>>
>> As you say there is the "interchange within the WG" issue that we've 
>> already discussed [1].
> 
> Why not exchanging diagrams (together with some serialization,
> e.g. in OWL)?

The issue is whether everyone needs a compatible diagramming tool in 
order to take the serialization and generate a modified diagram in order 
to be able to explain a point in email.

Clearly that can be made to work.

Other things being equal, though, a simple textual notation is easier 
for email-based working.

>
>> Second there is the issue that the point of this is for 
>> extensibility. 
>> In Sandro's "use a small OWL fragment with syntactic sugar" 
>> proposal we 
>> can exploit the extensibility of OWL. When an extension needs 
>> to add new 
>> productions to an existing abstract syntax node it just 
>> declares a new 
>> subclass. That new declaration can be in another self contained OWL 
>> model and merging the two models is both well defined and 
>> simple. It's 
>> not clear to me that either is true of merging in the UML case, but 
>> perhaps that's a limitation of my knowledge of UML.
> 
> Yes, it is. 

Understood.

> UML provides sophisticated ways of merging packages
> (similar to XMLS's possibilities with import and redefine).

XML Schema doesn't solve the problem either, it handles backward 
compatibility fine but forward compatibility is at least harder.

My point was simply that given one ontology then loading in a previously 
unknown extension which adds a few new subclasses is simple to the point 
of trivial.

What specific UML mechanisms are you referring to here? Presumably 
something more than MOF import.

>> Third, in Sandro's extensibility proposal [2] a RIF processor 
>> would go 
>> out and resolve the namespace of a unknown syntax element and 
>> obtain ... 
>> something. With the asn06-as-OWL approach that something could be or 
>> include OWL. Whereas the equivalent for UML, XMI, really 
>> seems to be a 
>> tool for diagram exchange rather than runtime data exchange [3].
> 
> No, XMI is a model, not a diagram, exchange language, so
> you have all of the language definition encoded.

Yes, apologies for the error. Since it was the XMI support of 
diagramming tools that failed for me in the past I had incorrectly 
pigeon-holed XMI in my ageing brain.

>> [By the way, how does "list of X" work in UML?]
> 
> I don't see for what this would be needed in an abstract syntax?

In the abstract syntax we do need to identify places where ordering is 
significant. Specially we want to say that the arguments to a 
RelationExpression or FunctionExpression are ordered so that positional 
notation is possible (unless that is pure (abstract) syntactic sugar on 
top of keyed syntax). In ASN.1 I believe you can distinguish between 
sets and sequences of subelements. I used the term "list" rather than 
"sequence" because that was the term used in the proposed asn06 (and to 
avoid confusion with RDF SEQ(uences)).

Dave

Received on Wednesday, 22 November 2006 09:50:55 UTC