RE: How to add the constraining facets of XML schema.

---- Original message ----
>Date: Sat, 18 Jun 2005 19:57:58 -0400
>From: "Geoff Chappell" <geoff@sover.net>  
[snip]

>I wonder about the need for complex types.

Seem as useful as any other type.

> On the one hand, I suppose
>they're nice from the perspective of being able to have islands of
>closed-worldness within your rdf (at least I always assume an implicit
>closed-worldness to xml, maybe that's not a universal belief?), 

I think of them more from the perspective of having structured aggregates. It 
would be nice if they could "call back out" to the rdf level, but oh well :)

> but on the
>other, I can imagine the debates about whether something should be
>represented atomically in an complex xml literal or be represented by its
>component parts in rdf.

Healthy debate, I would imagine. Seems a bit odd to close things off by fiat.

> At a minimum, I guess complex values might be nice
>for things like keeping units with values (and so avoid property names like
>lengthInInches).

Indeed.

[snip] 
>> It's making it *equivalent* that's hard. XML Schema is a beast, after all.
>> 
>> That being said, you could have a subset in RDF. That's what Jeff does.
>
>I guess I've really only thinking about a subset. I'll have to look at what
>he's done.

Personally, I don't see a lot of advantage in being able to do it in RDF syntax. I 
think it makes things harder, myself. I'm more mixed about inlining. I know a lot 
of people really like to have "one document" with all their stuff.

[snip]
>> >I'd hope that most reasoners would do the latter since it's just rdfs
>> >reasoning.
>> 
>> Well, *on datatypes*. That's not RDFS reasoning, IIRC. There are pitfalls
>> and
>> rooms for interpretation.
>
>Isn't it? 
>
>By:
> 	http://www.w3.org/TR/rdf-mt/#rulerdfs3 
>and 
>	http://www.w3.org/TR/rdf-mt/#rulerdfs10 
>
>seems you should be able to know that any value of myproperty is a
>xsd:integer (at least value-space-wise). 

You would also need to know that the constraints were consistent with the 
supertype, I think. There might be some trickiness with whether the subtype 
definition was complete.

> Admittedly things might be a bit
>murkier on the lexical-space side

On the value space side too. There are issues. I don't know all of them :)

> (and in practice, lack of support for
>literals as subjects might thwart some reasoners from making the right
>deduction). 

There would also, I imagine, be some issues with OWL DL.

>[...]
>> >Yeah, I don't think it does violate anything.
>> 
>> Well, I still don't know your expected behavior, exactly. You do expect a
>> clash if
>> the integer is outside the constrained range, right?
>
>Yes (plus hints to ui-generators). I've found facets are also handy on
>strings - e.g. when you're auto-translating a sql schema into an ontology
>and want to preserve the number of characters in a particular column/field.

Dervied types are handy, no question.

> [...]
>> > I chose
>> >not to go this route because my tools currently can't look inside xml
>> >literals during the reasoning process (except by treating them as
>> strings).
>> 
>> Oy! that sucks. XMLLiterals are much neglected :(
>> 
>> Which tool is that, btw?
>
>RDF Gateway [1]. To be clear, it supports xml literals fine wrt to
>serialization etc. and the system can certainly consume xml in a variety of
>ways. It just doesn't have built-ins to support inspecting xml literals as
>xml during the reasoning process (other than by treating them as strings
>with various string functions, regexp, etc.) -- so you can't easily create
>rules based upon parts of the xml literal. Might be worth adding some sort
>of xpath built-in to make these things available....

Yep.

Cheers,
Bijan.

Received on Sunday, 19 June 2005 01:21:16 UTC