Re: Classes and predicates as first class objects

Ian,

 I have to beg to differ with you on this point. This feature of RDF is 
one of its main virtues. Many systems I have helped build (including Cyc 
and tap) make extensive use of this feature. We have found it quite 
useful to be able to make statements both about classes and properties. 
This is why Tim Bray and I included this feature in [1].

 For example, TAP includes a category of properties corresponding to 
"abbreviation codes", i.e., airportCode, cityCode, partCode, etc. TAP 
also states many properties of classes (such as Musician) that help us 
represent their mappings to dmoz, etc. In Cyc, there is a whole big 
hierarchy of classes of classes which help make the distinction between 
substances and individuals (this work was derived from Montague).

 I submit that the experience to date in the usefulness of this feature 
suggests that RDF retain it.

 I do agree that this feature might prove fatal when combined with 
certain other features (as described by Peter's papers), but this is not 
always the case. As shown by Hayes & Menzel and Fikes & McGuinness, it 
is also fairly straightforward to construct first order logics that are 
quite comfortable with this feature. Further, as explained by Hayes and 
myself, the introduction of the concept of a reserved vocabulary, which 
keeps constructs such as first and last, which are essentially syntactic 
constructs, from showing up in the model, these problems can be avoided.

guha
 
[1] MCF using XML: http://www.w3.org/TR/NOTE-MCF-XML/


Ian Horrocks wrote:

>Guha,
>
>You chose to ignore my second point, so I will repeat it:
>
>    The ability to treat classes/predicates as arguments to other
>    predicates is of secondary importance. The crucial thing with RDF
>    is that it treats the vocabulary of the language itself as
>    standard classes/predicates that can be arguments to other
>    predicates. This is beyond the ability of almost all logics. It is
>    relatively harmless for a language as weak as RDF, but causes
>    fatal complications when more expressive power is added.
>
>Ian
>
>  
>

Received on Friday, 16 August 2002 11:16:11 UTC