W3C home > Mailing lists > Public > public-rule-workshop-discuss@w3.org > August 2005

Re: log:notIncludes (conclusion?)

From: Sandro Hawke <sandro@w3.org>
Date: Sat, 27 Aug 2005 16:35:30 -0400
To: Michael Kifer <kifer@cs.sunysb.edu>
Cc: public-rule-workshop-discuss@w3.org
Message-Id: <20050827203534.A05634EEC8@homer.w3.org>


> > > This is nonmonotonic.
> > 
> > I'm not sure I know the right words for this, but that stuff sitting
> > at that URL is considered immutable within one inference run.  cwm
> > makes no attempt to read it more than once; any change in it would go
> > along with an overall change in the state of the world.  A
> > long-running N3-based agent would have to start everything over each
> > delta-t (or cleverly act like it did).  Each time something observable
> > about the universe changes, we're talking about a new set of models
> > and interpretations.
> 
> Yes, but this is no different from the way Prolog (without assert/retract)
> works. You read the KB only once and then make NAF-based inferences. The KB
> doesn't change.
> 
> Nonmonotonicity doesn't mean that you are going to change the KB during the
> run.  It means that you make your inferences on a static KB with one set of
> models and observe the results. Then, some time later (after you are done
> with the inference) the KB changes. You reload your KB (or whatever the
> mechanism is for becoming aware of the change; it is more efficient in
> Prolog, of course) and recompute your models (again, assuming that the KB
> doesn't change during the inference). Then you observe the results again.
> Nonmonotonicity means that it is possible that when you observe the result
> the second time you may not find some inferred formulas that you saw when
> you observed the result the first time.

I think you left out the restriction that the only changes to the KB
made between runs are to add formulas.  Adding to a document which is
referred to via log:semantics and then log:notIncludes is not adding
to the KB.  Unfortunately, I'm not sure what exactly it is, logically
speaking.  Logically speaking, how do you talk about XSB's
file_exists/1 or cputime/1 ?


> > Going back to the definition of monotonicity you and DanC were using
> > [1], the stuff sitting at that URL isn't part of A or B.  Nothing is
> > entailed by that stuff at the URL.  To be slightly silly, adding a
> > formula to it is no more relevant to the notion of monotonicity than
> > is adding characters to the end of some constant symbol in an FOL
> > formula.
> 
> If nothing is inferred from formulas at URL then you are not dealing with
> semantics or logic. You are just manipulating lists and checking for their
> inclusion.

I think it's pretty fair to say N3/cwm is just manipulating lists and
checking for their inclusion with this stuff.

There are some other features which are more logical, like telling it
to automatically load the data from URIs it sees in certain contexts,
and log:definitiveDocument which says that all true atomic formulas
using a given predicate are in the data returned from a given web
address.   Perhaps that's what we should focus on.   (Or maybe we
should move back to what the charter can or should say about all
this.) 

> But I don't believe that nothing is entailed from those formulas.  What
> about your pharmacy example? You have two sets of rules sitting at two
> different sites that describe two different medicines. (Or whatever it
> was.) You pull those rules together and make inference that tells you that
> there is a possible side effect. So, these formulas at A and B are used as
> logical formulas and you are making inference from them.

That's not done with any of these features.  That's most simply done
by running cwm like

  bash$  cwm http://pharma-data-1.example.com http://pd2.example.com $Q

(where $Q will be inference-control and query parameters)

Of course that's conceptually the same as 

  bash$  cat http://pharma-data-1.example.com http://pd2.example.com | cwm $Q

except that of course cat doesn't handle URLs, and the information for
handling relative URIs in the content would be lost.

There's probably a way to use log:semantics and de-reify and assert
the retrieved data, but I'm not sure.   I should say that I'm
not arguing that cwm is better than runflora or anything, I'm just
hoping that we can identify the features and use cases causing some
apparent disagreements about the charter.  I think we're close.

     -- sandro
Received on Saturday, 27 August 2005 20:35:51 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:16:23 GMT