Re: bags of axioms

On Wednesday 31 October 2007, Pat Hayes wrote:
> >Hi,

> >Through having logical and physical URIs, the ontology elements are making
> >some sort of statement that they are a recognisable set of axioms, that
> > exist for some purpose, and so on.
>
> ? Why? This seems like wrong thinking to me. First, there's no
> distinction between 'logical and physical' URIs, so lets just talk
> about URIs. Think of a URI as a name. Giving a name to something is
> just making a statement that it can be referred to, is all. And that
> hardly amounts to saying anything about it. Google queries are
> evanescent things that have no location, but they are encoded as URIs.

Granted. This is perhaps more an issue of resolvable vs non-resolvable URIs, 
and the mechanisms used to turn logical URIs into physical URIs etc. that we 
all have to deal with in some way. Probably off-topic.

Regards the "Giving a name to something is just making a statement that it can 
be referred to" is exactly what we don't want to do - these bags of axioms 
shouldn't be refered to, they should just be moved about and absorbed by 
software agents. They are transient in existence.

> >The axioms we are moving about are either
> >sub-sets of the axioms within an ontology with a known logical URI, or are
> > a sub-set of the entailment of the axioms within an ontology with a known
> > logical URI.
>
> OK
>
> >They are not in themsevles a cohesive whole of axioms, but
> >rather more like incremental updates, deltas.
>
> But they are sets, right? So they are, for example, RDF graphs if
> written in OWL/RDF.

You can, but they don't really mean much without merging with other stuff. For 
example, we may have an ontology (T + A box) shared between both client and 
server, and then the server sends the client some OWL, which is meant to be 
considered in relation to the shared ontology, or that plus all the other 
deltas they have sent between them during the conversation.

>
> >  In particular, it makes no
> >sense at all to give them a physical URI.
>
> You mean, I think, a URI that anyone would expect to be able to
> resolve with HTTP GET to retrieve the set of axioms. OK.

Yeah. There's no location they can be got from - they are single-shot.

> >b) have no logical URI that identifies them, or any other identifier
> > smaller than themselves
>
> Don't have in fact, or can't have because you don't want to ever give
> them one? If the latter, why not?

It simply does not exist. These are anonymous bags of axioms. They do not 
exist outside of the communication transaction. They should never be refered 
to outside of the communication transaction.

> >c) are a sub-set of the axioms entailed by some other ontology, which does
> >have a logical URI
> >
> >Is there some other top-level element other than Ontology that better
> > fulfills these needs?
>
> Whats wrong with Ontology? Nothing you have said is inconsistent with
> these things being in the class Ontology.

Well, OWL/XML requires a URI attribute on the ontology element for a start. 
Also, all of the tools we have looked at require the ontology to have a URI.

> They don't HAVE to have a 
> URI, you could refer to them with Bnodes. But in fact it is a very
> low-cost business to construct a unique URI, rather like gensym in
> GOFLISP. Use, say, the base URI of the derived-from ontology and
> append the real time in seconds the thing was streamed. Or whatever.
> Of course if you try to GET or owl:import that URI things will break.
> So, don't.

In practice, this is close to what we do actually do. However, it feels ugly 
and dirty to be minting URIs that serve no purpose other than to get a file 
to validate, and which no software should ever use in any way for a bag of 
axioms that where only packaged together for transmission purposes.

> >If not, is there some standard way to indicate to tools that the ontology
> > has no physical or logical URI, but is a) a sub-set of the axioms in
> > another ontology or b) the axioms entailed by another ontology or c) a
> > sub-set of the axioms entailed by another ontology?
>
> Not a standard, but you could invent this subclass of Ontology if you
> want, call them EvanescentOntologies or something.

Ah, yet more non-standard solutions :) Don't we all love forks.

>
> Pat

Matthew

Received on Thursday, 1 November 2007 16:16:01 UTC