Re: A bit of running code for "Tim's View"

>I proposed a kind of "Hello World" for Tim's view of how URIs should
>work in RDF, and he modified cwm to support it.  I take this as pretty
>good evidence of what his view is.  :-)

very nice.

>The test is:
>
>     1.  File myDingo.n3 says:
>
>           @prefix bio: <http://www.w3.org/2000/10/swap/test/ferrell/biology#>.
>           :Donny a bio:Dingo.
>
>     2.  If you follow the bio:Dingo link, you will learn that every
>         bio:Dingo is a bio:Dog.  Try it.  Bio says it using
>         ferrell:subClassOf.  [Ferrell is another name for OWL, but I
>         don't want the machine to know that.  Ferrell is the name of a
>         great horned owl my wife used to babysit.]  If you follow the
>         ferrell:subClassOf URI, you'll see a doc:rules link which
>         points to some Horn rules about what ferrell:subClassOf mean.
>	Some of this (like doc:rules and log:implies) is only defined
>         for humans, but enough is machine-readable that for step 3 to
>         work.
>
>     3.  Run "cwm --closure=por myDingo.n3 --think" and it will
>         output (among other things):
>
>           :Donny a bio:Dog.
>
>While "--closure=poEr" looks pretty obscure, I believe Tim thinks it
>should be a very common way to do things.  From the docs:
>
>    Closure flags are set to cause the working formula to be
>    automatically expanded to the closure under the operation of
>    looking up:

Looking up? Or drawing conclusions??

>
>     s   the subject of a statement added
>     p   the predicate of a statement added
>     o   the object of a statement added
>     t   the object of an rdf:type statement added
>     i   any owl:imports documents

? Do you mean, any owl:imports statements? (In what document?)

>     r   any doc:rules documents
>

Well, that is very sweet, but....

>My proposed generalization would be:
>
>    Applications which use RDF *should* offer users options for
>    following links in the RDF and recursively including content.  A
>    good default is to follow predicates, objects, and (for rule
>    systems) doc:rules.
>
>This kind of "Applications ... *should*" is more along the lines of
>the HTML spec (or protocol specs in general) than the existing RDF
>specs, but it seems to me the best approach here.

... if that really is RFC 2119 language then I think this is WAY too 
strong at the present stage to put into anything as wide-ranging and 
authoritative as a TAG architectural statement.

>Following links seems to me very much like a reasoning step.

NO!!  Please don't say that. Why is it anything AT ALL like 
reasoning? This is like saying taking a book of a shelf is a form of 
reading.

What might be good is to say that following a link is a way to gain 
information, and so in a sense is reasoning (where 'gain' means 'make 
explicit'), but that generalization is too valuable (and at the 
current stage of the game, too fragile) to waste.

>  Guiding
>software which can follow links around the web is a lot like guiding a
>resolution theorem prover: you only terminate if you get lucky, you
>try to trim whole branches, you try to tell the machine where to focus
>its search.
>
>Of course following links is probably not a *sound* reasoning step,

Its not a reasoning step at all; soundness doesn't enter into it. If 
you wanted to say, instead, that a good notion of justification may 
use the act of following a trusted link as being itself a form of 
*justification*, that would be different: I would agree with that 
immediately.  That's like saying, I believe this because Joe told me 
it, and he never lies.  (Which assertion itself might be supported by 
reasoning, of course, and require justification, and so on..) So we 
need new notions of justification which generalize and extend the 
current notions of purely logical reasoning: a new kind of proof 
theory. Right, we need that. But that's a different thing to say.

>but I think it was Bijan who first told me soundness was vastly
>overrated.

Soundness *in reasoning* is not over-rated. Soundness more generally 
has yet to be adequately defined.

>(Will this finally divide Peter and Bijan's positions?)  I
>imagine this being addressed by a user interface which keeps
>justifications very handy, and which allows some configuration about
>which sources (and reasoning steps) to trust.

That would certainly be very nice and probably essential reasonably 
soon. But I think there is still some basic work to do before we can 
get there.

Pat


-- 
---------------------------------------------------------------------
IHMC	(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.	(850)202 4416   office
Pensacola			(850)202 4440   fax
FL 32501			(850)291 0667    cell
phayes@ihmc.us       http://www.ihmc.us/users/phayes

Received on Wednesday, 8 October 2003 15:55:25 UTC