Re: Hidden triples and self-description

Mark Baker wrote:

>First, let me define a "hidden triple" to be a triple which is not
>extractable from a document by an RDF processor.  A trivial example is
>this; 
>
>:myProperty rdfs:domain :myClass .
>:some-object :myProperty :some-other-object .
>
>where one hidden triple (are there more?) would be;
>
>:some-object rdf:type :myClass .
>  
>
That would be an _entailment_ licensed by either the RDF or OWL model 
theories see: http://www.w3.org/TR/owl-test/#testEntailment

>As a REST fan, I'm always on the lookout for things that appear
>non-RESTful, and I believe this may be one, as that triple is not
>visible to either intermediaries nor the recipient, unless they also
>know RDF Schema (this isn't RDF Schema specific, of course).
>  
>
Well yes I suppose but similarly an HTML document cannot be displayed by 
an intermediary or agent that doesn't know HTML. Indeed the RDF and OWL 
model theories are published in well known places, so isn't it sort of 
obvious that if you are going to use the RDF or OWL model theories to 
derive entailments, that you'd have to know the rules of entailment?

>Is this a problem?
>
>>From a REST POV, which prescribes self-descriptive messages, I think
>it is.  It seems to me that if there's a disconnect between what the
>sender intends to communicate (which may or may not include the hidden
>triple), and what the recipient or intermediaries understands them to
>be communicating, then there is a failure to communicate.
>  
>
Huh? What about text/plain? What about text/html? .... there are all 
sorts of reasons why there might be a disconnect between what such 
senders and recipients understand the messages to _mean_ ... gosh at 
least RDF and OWL have rather precise model theories that go a long way 
toward codifying the rules. What you are worried about is _hardly_ an 
issue with the semantic web, which is vastly better than the current Web 
in nailing down such ambiguities of communication.

>Using that example above, one way to make it self-descriptive, such that
>the hidden triple is not hidden, would be to use an RDF Schema specific
>media type.  There are pros and cons to this approach, and I have some
>ideas that could make this quite attractive - but I'll save them for
>another time.
>  
>
well there is an RDF specific media type application/rdf+xml which 
delegates semantics to the RDF model theory via the RDF specs ... this 
is all in the media type registration. RDF Schema shares a model theory 
with RDF. Now I've argued that OWL ought have its own media type beyond 
RDF but there are pros and cons to this approach (you can read through 
the www-webont archives if you are motivated) and most people on the WG 
weren't motivated to define an OWL specific media type.

>Another would be if it was the intent of the sender to not attempt
>to communicate the hidden triple.  This, I'm quite fuzzy about; what
>does it mean that a sender wants to communicate, versus not, a hidden
>triple?  Does it suffice to say that no sender should ever intend to
>send a hidden triple?
>  
>
Well you can always send RDF as application/xml in which case you aren't 
defining any triples at all... just sending RDF/XML _as XML_. In any 
case the idea is that if you use rdfs:domain or rdfs:range, the 
_meaning_ of these predicates is derived from their corresponding URIs 
e.g. look at the RDFS namespace document for rdf:ID="domain" ... this is 
a current TAG issue "meaning of RDF predicates"

>So, is this well worn territory that I haven't been exposed to, or
>perhaps something that merits a closer look by those who know a whole
>lot more about Semantic Web underpinnings than I do?
>
>  
>
I don't have a REST issue with this. I would like some specs to make the 
connection between the meaning of a predicate and its URI more explicit 
i.e. provide a rule such that the meaning of an RDF predicate (such as 
rdfs:domain) is determined by dereferencing its URIref using 
application/rdf+xml and processing the definition according to what is 
found ... if it points to a triple whose subject is "rdfs:Class" or 
"rdf:Property" then process according to the RDF(S) model theory, and if 
it points to "owl:Class" or "owl:ObjectProperty" then according to the 
OWL model theory... there remain a few issues but that's how I see it. 
At some point when we are dealing with the meaning of media type 
specific messages we need to delegate to the media type registration. 
The Semantic Web is all about doing this in an automatic fashion (e.g. 
each media type would (ideally) have its own model theory ... and they 
might all fall within the L-base framework).

my 2p.

Jonathan

Received on Saturday, 13 September 2003 13:51:53 UTC