Re: TEST: maybe if/then is clearer than consistency? (more on sameGuy)

On March 7, Dan Connolly writes:
> I propsed tests consisting of a file of
> premises and a file of conclusions...
> an "if" file and a "then" file.
> 
> Ian, you pointed out that the usual way inference
> systems do that sort of thing is to
> negate the conclusion, assert the whole
> thing, and ask if the result is consistent.

Actually, now that I look at them, most of my tests check if some
concept is inconsistent (rather than the whole KB being inconsistent).

> First question: does that always work?
> is DAML+OIL closed under negation? How
> do I negate a simple
> 
> 	<page1> dc:title "ABC".
> aka
> 	(dc:title page1 "ABC")
> fact?
> 
> I guess the result looks like:
> 	(rdf:type page1
>           (complementOf
>              (onProperty dc:title hasValue "ABC")))
> i.e. page1 is not in the class of things
> that have "ABC" as a value of the dc:title property.
> 
> I still don't know if the whole language
> is closed. Ian?

It is a condition of reasoning with Datatypes that they be closed
under negation, otherwise the above kind of thing can't be decided
(the formal stuff on datatypes says this somewhere). I.e., the
datatype reasoner/black-box has to be able to decide if a there is a
single value that will satisfy a set of (possibly negated) datatypes
and values.

> But in any case... the result is pretty
> contorted. I'm working on taking the
> conclusion of the sameGuy test and negating
> it, and the tutorial value of the example
> is going right out the window.
> 
> And the resulting test relies on not
> just UnambiguousProperty, but also
> onProperty/hasValue, intersectionOf,
> lists, and all sorts of other features
> that I didn't intend to test; i.e.
> that my use case doesn't rely on.
> 
> So maybe there's a role for if/then
> tests after all.

You could be right - it is late here and I am too tired to think about
it any more. However, before collapsing into my bed I did get round to
putting some test stuff at:

http://www.cs.man.ac.uk/~horrocks/Private/DAML/Tests/

There you will find:

   d+o-tests.lisp          07-Mar-2002 23:21    18k  

Various tests in lisp syntax, but hopefully self explanatory enough
for you to translate. Some of them are taken from papers, some reflect
bugs discovered (and fixed) in FaCT, some are just attempts to break
the system, and some test efficiency of reasoning with cardinalities.

All of the tests define small ontologies and then check if various
concepts are inconsistent.

As a starter for 10 I converted 3 of the tests into DAML+OIL using OilEd:

   heinsohn1.daml          07-Mar-2002 23:20     9k  
   shiq-t5f.daml           07-Mar-2002 23:20     6k  
   shiq-t6f.daml           07-Mar-2002 23:20     8k  

Enjoy.

Ian

> 
> -- 
> Dan Connolly, W3C http://www.w3.org/People/Connolly/
>

Received on Thursday, 7 March 2002 18:34:04 UTC