RE: The mentography of reification

>  > -----Original Message-----
>>  From: www-rdf-logic-request@w3.org
>>  [mailto:www-rdf-logic-request@w3.org]On Behalf Of Pat Hayes
>>  Sent: Wednesday, October 24, 2001 8:32 PM
>>  To: Geoff Chappell
>>  Cc: www-rdf-logic@w3.org
>>  Subject: Re: The mentography of reification
>>
>>
>>  >My understanding is that the triple can be thought of as defining a
>>  >particular arc in a graph. That nodes and arcs have identities
>>  (locations on
>>  >a page, position in memory, or whatever) and labels.  That with the
>>  >restriction that no two nodes can have the same label,
>>
>>  We may want to relax this slightly for literals; but otherwise, yes.
>>
>>  >  we can uniquely
>>  >identify a node by its label. That with the restriction that duplicate
>>  >triples can not exist, we can uniquely identify an arc by the nodes it
>>  >connects (in order) and the label on the arc. (Nodes, I guess,
>>  are asserted
>>  >into existence by their use in describing an arc?)
>>  >
>>  >Taking that view, I'd always envisioned that a nested or reified triple
>>  >would be shown on a graph as arcs originating or terminating on
>>  arcs (though
>>  >I don't know about the validity of that in graph-speak).
>>
>>  It isn't good graph-speak, and it isn't correct RDF either, so don't
>>  think of it that way, I would suggest.
>>
>
>Thanks for the suggestion :) But I can't help thinking there's something
>clarifiying about that way of looking at it.

Well, I think it is less than clarifying, though it may be possible 
to clarify it with some work.

First, I'm not sure what you mean by 'arc'. If this is a piece of an 
RDF graph (drawn in ascii-art):

nodeaaa ------ edgeccc------> nodebbb

is that an arc? Or is THIS an arc?:

------ edgeccc------>

ie does your 'arc' have three labels or one? If the answer is three, 
this is what is called a 'triple', and its the simplest reifiable 
statement in RDF. If this is what you mean, then indeed reification 
could be indicated by some such convention of arcs pointing to arcs, 
I agree (though I can see no utility for arcs coming FROM arcs). I 
took you to be talking about the other idea, and I can see no real 
utility for that one.

We have to be very careful not to think that this convention was 
indicating a property with a *property* as value, however, since the 
same property label might (usually will) be on other arcs as well as 
this one. There isn't any 'thing' in the graph corresponding to a 
property.

Also, we have to take care not to interpret these arcs-to-arcs in the 
same way that node-to-node arcs are interpreted.

All this can be done, but we need to be careful.


>To make it a valid picture I
>suppose it requires elevating labeled arcs to nodes themselves with arcs
>then just identifying the binary connections between nodes. Couldn't that be
>considered an alternative visualization/representation?

Well, that's pretty much what current reification does, right?

>  I realize that
>deviates from the normal view but surely it captures the essence of rdf -
>its binary predicate-ness and readily allows for (future) deeper structures
>(nested triples, lists).

That is what worries me. Considered purely as a graphical convention, 
this could mean all sorts of things. If we use it for reification, 
then we don't want to also be using it for all these other things as 
well, or confusion will reign. In particular, we certainly don't want 
to be using it for arbitrary lists or nested triples.

>It probably requires giving nodes an intrinsic type
>but that seems beneficial for other reasons as well.
>
>I realize this falls easily under the "not rdf" category - as everything
>that rdf might become does. But surely what gets defined today tries to
>anticipate the rdf of tomorrow. How does the current graph conceptualization
>of rdf handle deeper structures?

The current conceptualization of RDF *doesn't* handle deeper 
structures. Let me return to my original question. Why should we 
care? There are hundreds of alternative notations for expressing 
complex nested structures, many of them with a long history of use in 
programming languages, databases, industrial applications of one kind 
or another, etc. . Their various powers and limitations are 
well-known. Simple binary graphs, as in RDF, aren't among the leading 
contenders for a general-purpose structure notation. If you want one 
that is, pluck one off the shelf. LISP Sexpressions, labelled dags, 
XML, nested semantic networks, almost any recursive functional 
grammar, all kinds of box-and-arrow notations, Piercian assertion 
graphs, CGs, FOL, Prolog, hypergraphs, KIF, BNF, even simple nested 
bracket structures can all get the job done one way or another. 
Labelled binary graphs cannot.

Pat Hayes
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Friday, 26 October 2001 17:17:03 UTC