RE: Annotation properties with range and other rdf:types

At 1:32 PM -0800 12/3/03, Holger Knublauch wrote:
>>  Holger, I must admit some confusion - if you know the types and
>>  properties of these things, why do you want to make them annotations?
>>  If you goal is to have metaclasses (I.e. classes with properties) you
>>  can certainly use OWL Full, it was designed to allow that.
>
>Yes, I could use OWL Full, but I think the feature of (typed)
>annotation properties is of such a great importance to ontology
>designers that it should also work for OWL DL.  While I could
>delete all annotations temporarily before I send the ontology to the
>classifier, the information will still be stored in the ontology,
>making the file difficult to share on the Web.  My apologies if
>I lack the technical background of the OWL language theory that
>the WebOnt group has, but what is the problem with allowing it
>in OWL DL?
>
>Our typical use case is to allow the definition of metaclasses and
>property metaclasses which hold extra information.  For example,
>a property metaclass could introduce an extra property "unit"
>or a class could be annotated with an "author" and "change date".
>This information is completely irrelevant to the classifier,
>and could be easily ignored if the property is flagged as an
>annotation property.  I don't see however, why this excludes the
>use of range information on the annotation property.
>
>
>>    So, in essence, you cannot have it both ways -- if you want to live
>>  in DL, you need to accept its restrictions.  If you don't want these
>>  restrictions, you should use OWL Full -- afraid that logic will not
>>  let us have our cake and eat it too (believe me, the WG tried hard to
>>  find a way to do that :->)
>
>My impression is that, in this aspect, OWL has been designed
>to fit the needs of a classifier, and not from a user's or tool
>perspective, but you know more than I do.
>

Yes, that is exactly right -  the OWL DL profile is defined with 
decidability and completeness in mind - you cannot have those 
gaurantees without some restrictions.  Owl Full is defined to be more 
compatible with the wild, wooly RDF world --more seriously, for 
applications where the logical guarantees can be traded off with 
other issues -- performance, expressivity, database compatible, etc...

>
>>    For what you are suggesting doing, why can you not just use RDF
>>  comments, etc. or to define things directly in OWL?
>
>Because from a user's perspective annotation p's should appear
>just like other properties - he or she should be able to specify
>extra information (e.g. range) on the property so that it is
>later easier to enter values for the property.

So  this means you need to be in OWL Full, which is fine - nothing 
wrong with that, and many nice reasoners are being written for OWL 
Full - but you won't get certain guarantees.  For what it is worth, 
the projects in my research group include some which use Lite and DL 
and others which use Full.  All of our "web presence" (the 
http://owl.mindswap.org web site for example) currently uses special 
purpose reasoners that only do some small sets of inferences (and 
don't need to do classification) and those work with relatively 
arbitrary mixes of RDF and OWL (i.e. OWL Full) -- there are other 
projects where we use Pellet, our OWL DL reasoner, and we must make 
sure the documents are more structured.   There seems to be some myth 
that somehow DL is better than Full or something like that, but if 
you look at our documents you see it is all OWL -- you just get to 
choose what you need.
>

Hope that helps -- and let me encourage you to allow Protege users to 
create OWL Full documents when they want to -- there's no reason for 
it to insist everything be in Lite or DL -- however, would be nice if 
those users who want to stay in those profiles could - perhaps 
putting the output through Pellet or WonderWeb or such and saying 
"You have created an OWL Full document.  To make it Owl Lite you 
could <xxx>" - doing that with full generality is probably hard or 
impossible, but catching most of the main errors isn't that bad -- 
several of the implemented systems already do some of this


>Holger

-- 
Professor James Hendler			  http://www.cs.umd.edu/users/hendler
Director, Semantic Web and Agent Technologies	  301-405-2696
Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
Univ of Maryland, College Park, MD 20742	  240-277-3388 (Cell)

Received on Wednesday, 3 December 2003 18:27:50 UTC