Re: regarding rdfms-identity-anon-resources

Jonathan Borden wrote:

<snip />

> Let's define roundtrip then. The simple definition of roundtripping an XML
> document is that I get _exactly_ the same XML document out that I put in. To
> do this you need to either store the XML document as text or build a full
> grove. Not many applications can roundtrip an XML document due to this
> stringent requirement.
> Perhaps you mean that there is some information in the XML document you wish
> to preserve? What? Define what and you subset the XML grove. The XML Infoset
> is what people think of as a 'common' base, but the XML Infoset doesn't
> preserve things like: order of attributes etc. Do you wish to preserve the
> order of elements? Whether an arc was derived from an element or attribute?
> None of this is currently represented within the RDF model.

This is a very useful direction and I really hope that it will be one
of the focuses of the working group. The concept of an RDF infoset has
been bandied around at different times. Without getting into the
particulars of what the RDF infoset is, I would claim that it is
neither the M&S concept of a model or the XML concept of an infoset.
It is closer to the Hytime concept of a hyper-grove in that the meat
of RDF is related to the relationship between documents rather than
the internal structure of a specific document. 

There is currently information lost on the trip from RDF/XML ->
triples. There is also information lost (or the inability to express
it) on the trip from triples -> RDF/XML. Assuming that both of these
trips were possible losslessy, the next step would be to define the
round-trip that would not change whatever was considered the relavent
aspects of the information. 

As to the specifics of what should or shouldn't be maintained on the
RDF/XML -> triples trip, I certainly have some opinions. I think that
anything that the M&S explicitly calls out as being representable in
the XML syntax should be preserved in the triple representation. This
includes things like lang, anonymous resources, reified vs. ground
statements and bags used as context.

I also happen to think that the distributive mechanisms that are
described in the M&S like aboutEach, aboutEachPrefix should be
represented in the triples, but unfortunately, this is left as
implementation dependant in the M&S. 

> > >         The issue of child seqs being invalid XML names is by
> > design! If the
> > > resource _had_ a valid XML name it wouldn't be anonymous would it?
> >
> > If anonymity was a first class concept, ala a special variable type,
> > then you wouldn't have to conflate lack of a name with it being
> > anonymous.
> >
> I am defining 'anonymous' operationally and since anonymity _isn't_ a first
> class concept then lack of a 'known name' _is equated_ with being anonymous.
> This is precisely why I don't want to get into an abstract discussion of
> anonymity -- I see no end in sight. If you can first persuade me that there
> is a specific and practical consequence of defining 'anonymous' this way or
> that way then fine. So far I see the questions simply as:
> 1) Can you assign this resource a URI reference for the purposes of parsing
> an XML document as RDF?
> 2) Was the node given an explicit name? If so, what was it.

I agree with this operational approach. If it seems reasonable to use
childseq fragment identifiers as the signaling mechanism then so be
it. It would certainly be better than the current morass. I can see
the use of childseq as identifiers as also being useful to identify
literals. This assumes that you want to use URIref to refer to
literals rather than embedding them directly in the triple as is done

> Parenthetically, child sequences _are_ useful for reconstructing element
> order in documents because they are based on element order -- to the extent
> that this might be helpful in roundtripping. On the other hand, at the
> present time we can't even generally serialize an RDF model using the RDF
> syntax so we are far far away from being able to roundtrip.

I don't see recreating the element order as high priority. I don't
think we are that far from being able to serialize an RDF model to
XML. OTOH, I haven't focused much attention on it. When I have thought
about it, I have assumed that the model would be enhanced (on the
in-trip) with explicit metadata that would make the serialization of
the model (on the out-trip) straightforward. I.e. explicit information
on the mapping between a particular type or properti URIref to the
equivalent qname.
> -Jonathan



Received on Saturday, 10 March 2001 17:14:34 UTC