Re: nodes and node labels [was New document: revised version for WG review]

Jeremy Carroll wrote:

> 
> I wanted to draw the WG attention, particularly Pat, to a mathematical issue
> with the graph syntax.
> 
> I do not believe that this has any substantive content, but is merely
> aesthetic!



Aesthetic possibly, but it also has implications on apparent 
consistency.  I'm explaining some of this stuff in the Primer as well, 
and it isn't going to help the understandability of our specs if we 
appear to be inconsistent in how we talk about truly basic stuff like 
the various parts of the graph (and whether things are "labels" or not). 
  I don't care if the difference is trivial in "substantive content", 
the mere *appearance* of a difference creates a problem.  I think having 
such differences ought to be rigorously avoided, unless it's absolutely 
God-awful necessary.  In case you're wondering, the Primer currently 
uses the description from the "older" version of the model theory;  but 
I'll be happy to change it to gain consistency, once I make sure what I 
need to be consistent with!  (my default is the model theory)


> 
> 
> In the very first version of the model theory, the RDF graph was described
> as having nodes some of which had labels, and the labels were URI refs or
> strings.
> 
> In the most recent version, some of the nodes are URI refs and some of the
> nodes are strings (and none of the nodes are labelled).



A couple of minor questions about this:

a.  if, in the current model theory, some nodes *are* URIrefs, rather 
than being *labeled with* URIrefs, why shouldn't arcs *be* URIrefs, 
rather than being *labeled with* URI refs?

b.  if some nodes are URIrefs and some are strings (and we say nothing 
about them being labeled any more), why are we so insistent to say that 
blank nodes are unlabeled, and that node identifiers are *not* labels? 
Apparently, at the moment, no nodes have labels.  (This node identifier 
business, by the way, is one of the things I'm trying to explain in the 
Primer that I don't think is helped by inconsistency among the documents).

Pat?


> 
> In the new document draft, we have reverted to the earlier version using
> explicit node labels.
> 
> The motivations for this include:
> - tidying a graph is an explicit operation rather than implicit by
> mathematical construction
> - it easier to modify the exact tidyness specification (if the WG changes
> its mind about whether literals are tidy or not - uriref nodes don't seem in
> doubt).
> - implementations will almost all use nodes with explicit labels.
> 
> 
> The first point is the decisive one. From a mathematical point of view Pat's
> latest model theory treatment in which the URI refs and strings *are* the
> nodes is extremely elegant. Tidiness just falls out and no text needs to be
> spent on it.
> However, in practice any implementor needs to be aware of tidiness as an
> operation which has to be coded, and any user needs some understanding of
> tidiness. Thus, I felt that Pat's earlier treatment, while a little more
> clunky, is clearer for our intended audience: implementators, users, web
> architects - rather than mathematicians.
> 
> 
> Although I raise this issue now, I would hope that any debate, if debate is
> needed, can be postponed until after the first WD. I don't think this is a
> crucial issue either way.
> 



Well, I think debate is needed, not so much about which explanation is 
better, but about whether we ought to use one, and not two (or, heaven 
help us, more than two!).  And I think *this* debate ought to happen 
ASAP.  It seems to me that if using the basic concepts from the model 
theory causes problems for non-mathematicians, we need to explain things 
better, not introduce apparent inconsistencies about whether nodes *are* 
URIs or are labeled by them (and I suspect that the non-mathematicians 
will be the first to notice, and have a problem with such inconsistencies).

--Frank

PS:  You guys did say this new document was going to *help* with 
something, right?  :-)



-- 
Frank Manola                   The MITRE Corporation
202 Burlington Road, MS A345   Bedford, MA 01730-1420
mailto:fmanola@mitre.org       voice: 781-271-8147   FAX: 781-271-875

Received on Monday, 29 July 2002 19:59:28 UTC