Re: names, URIs and ontologies

Pat,

You asked for comments, so here we go.

I agree with your description of the problems around the naming of
concepts.
IMO one of the reasons why we have difficulties with them is that we
(humans) take often the name of a concept as the concept itself. For me
the name of a thing is only one of the descriptive elements of the
concept it stands for. 
This ambiguity can be found in RDF too, and that's one of things I'm not
very happy with in RDF. In RDF when a name is defined by an URI, what
the URI is in fact describing? IMO it will often be a communication /
descriptive element of the concept it stands for not the concept itself.
So a name is not an identifier of a concept. If you have a look, in RDF
terms, of how Boston (US) is "defined" you will end up with a lot of
URI's, all of them will use the term "Boston" and other information
about Boston will be present in "close" URI's (e.g. the same document).
So the concept Boston (US) as a whole with be scattered throughout a
whole set of URI's, and if you want to refer to it you will have the
need to refer to at least some of them (not one) in order to be sure to
be able to hook in to this network describing "Boston (US)". 
The reification mechanism of RDF allows the generation of an abstract
concept. The use of this possibility is though limited to the high order
statements (statements about statements) but doesn't, as far as I can
see, allow the direct identifying of the reified subject.
If reification for identification would be possible in RDF, the whole
problem will be solved.
This concept reification could than be considered a new layer in the WWW
perhaps in a way done for DNS (<smile>something like CIL Concept
Identification Layer</smile>).
I did something similar in Notion System (http://www.notionsystem.com).
This is a KB and at the same time something very close to a semantic
network, I developed almost 10 years ago. Speaking in terms of RDF, the
reified concepts (notions) are identified by a unique ID (which is a
URI) and are described by a whole set of names (different languages,
different domains, etc.) and by a set of relationships it has with other
reified concepts (notions). You can have a look at some of the sample
data (e.g. my concept at
http://users.raket.nl/rapoell/NSWebSite/Standard/Public/6A6A7269/6A6A707A/23686F6D/61616167.htm)
in order to have an idea how this done.
One of the possibilities of the implementation of the identification
layer is explained on the informational pages on this site, which
describes also how availability of it can be assured (through cloning
mechanisms).
One might say that the ID in this layer is just another URI for the
concept. This is not completely wrong, but taking in account practical
considerations like the need for permanent availability (through time
and place) I prefer to see it as something slightly different. Within
the concept of this extra layer you will not use "hardcoded" URI's but a
reference (ID) from which the actual best available real URI can be
delivered by the layer (perhaps we should call it an indirect URI).

Something that is not clearly stated all the time, is the fact that
Human-Machine-Human communication and Human-Machine-Machine
communication need different elements in order to provoke similar
perceptions on both sides of the communication chain. So concepts should
be able to deliver identification / recognition elements depending on
target type (human or agent). For machine targets (agents) perfectly
unreadable ID's are perfect. For human targets you will need a more
descriptive way (names, pictures, contexts, relationships, etc.). This
is true as well for the concepts themselves as for the relationships
between them (and we are talking about semantics here).

<Pat Hayes>
Anyway, I'd be interested in any comments.
</Pat Hayes>
Me too.

Ronald Poell

Consultant Knowledge Management
TNO - Netherlands
http://www.tno.nl
http://www.notionsystem.com

poell@fel.tno.nl
rapoell@notionsystem.com

Received on Saturday, 28 October 2000 03:52:45 UTC