Re: Repeated properties versus Bags

Ralph R. Swick wrote:

> We considered doing exactly this; in fact, we started with a
> design choice to disallow repeated properties.  But some of
> our client applications felt that Bags were too cumbersome
> for some common situations such as a work with multiple authors.

That sounds as a quite weak reason.

If the need is real it might be easily answered by defining a simple
transformation rule
so that:

<author>Karl Mark</author>
<author>Groucho Mark</author>

is transformed to:

<author>
 <rdf:Bag>
  <rdf:li>Karl Mark</rdf:li>
  <rdf:li>Groucho Mark</rdf:li>
 </rdf:Bag>
</author>
 

> A second reason for not disallowing repeated properties in
> the data model is the observation that there can in general
> be multiple Descriptions of the same resource.  We do not
> wish to hard-wire into the core of RDF a single mechanism
> for conflict resolution (do we discard some properties if
> they are repeated and if so, which one(s)?  do we discard
> the entire Description?)
> 
> If, for example, two Descriptions give separate prices for
> an item and one Description is further described by a
> property saying "valid 1998-07-28 only", there is no useful
> way the core RDF can know enough semantics to recognize
> that there is no logical conflict.

I'm not sure that this is a valid reason.

Description merging it's a problem that can probably be solved only at
the application level as it
has to do with the precise semantic of the involved descriptions.

Repeated properties is no solution. 
In the general case I think that it would be definitly "criminal" to
produce a new description by simply adding the properties of a number of
partial description as this could probably alter the semantic in
unexpected ways (would you trust an RDF system running in a nuclear
power station near you working that way?)

As description merging it's a real problem it should be treated as such.

Ideally, in the context of RDF schema, a way of specifing the merging
rules should be specified.

These rules might be specified as easily with respect to a repeated or
unrepeated properties RDF definition so
this is clearly a separate issue.

Regards

-- 
Pasqualino "Titto" Assini  ---  assini@kamus.it  
Kamus Internet Consulting  ---  http://www.kamus.it/

Received on Thursday, 30 July 1998 08:03:15 UTC