Re: Self-descriptive assertions

On Mar 25, 2004, at 11:20 PM, Mark Baker wrote:

> On Thu, Mar 25, 2004 at 02:17:02PM -0500, Thomas B. Passin wrote:
>>> True, but why is that bad?  I believe it's because - as I said above 
>>> -
>>> that the recipient would believe that the sender is trying to
>>> communicate the graph.
>>>
>>
>> But the whole point is that the sender has no say in how the recipient
>> chooses to process the representation.
>
> Right ...
>
>>  Maybe I want to get an example
>> of application/rdf+xml to illustrate an article I am writing.  Maybe I
>> want to extract certain information using xslt and never need to form
>> triples.  Maybe I want to apply some non-RDF processing as I build the
>> graph.  Maybe I want to somehow canonicalize the data and end up with 
>> a
>> different (but we hope equivalent) one.  Maybe I have a quad system 
>> and
>> want to load the RDF into quad statements.
>>
>> Or maybe I want to do what you think I ought to do.
>>
>> So the only area we can have a reasonable hope of working with here is
>> what the _sender_ may have wanted to communicate beyond the actual 
>> data
>> contained in the application/rdf+xml representation.
>
> Yes, exactly.  That's what I mean by "communicate the graph".  If I
> use text/plain, I'm not communicating the graph, because the message
> doesn't include any information would inform a recipient that the
> message semantics depend on the RDF specs.

This is a total non-starter, IMHO. It doesn't work either way: I don't 
necessarily "communicate the graph" by using application/rdf+xml (think 
about a tutorial on the hidiousity of RDF/XML), nor do I necessarily 
*fail* to "communicate the graph" by not using it.

Note that by your model, there's no way to "communicate" an owl 
ontology (since we don't have an owl mimetype).

> In both cases (application/rdf+xml and text/plain) there is no
> requirement that the recipient extract the graph, as that would be a
> requirement on processing.

I don't want to put a yicky requirement on senders either :)

>> Now, the sender _may_ be wanting you to think "Yea, verily, this
>> information is true, and I _am_ expecting everyone to apply RDF
>> interpretation rules to it", but there are many other possibilities.
>
> Yup.

Hence you can't infer things *either way*. I.e., you can't assume that 
the sender *didn't* want her text/plain to communicate the graph.

Mimetypes don't tell use a lot about content, actually. Imagine a gif 
of an rdf graph. That might well communicate the graph! Why not? (e.g., 
I have an ocr program)

Cheers,
Bijan Parsia.

Received on Friday, 26 March 2004 11:25:59 UTC