Re: TEST: please do Nothing

Ian - let me address this one from the chair perspective.  The tests 
Jeremy proposes below (i.e. basic tests for all language features) 
are necessary to our moving to completion of the Recommendation 
process.  The production of the sort of tests you propose are 
definitely desirable, but not necessary.  If time permits I would 
love to see us have a comprehensive test set by which various tools 
and reasoners can be tested.  If you or others have the extra time to 
prepare such, please submit them.  However, I do not believe the W3C 
management/AC will believe we can do an abbreviated Candidate 
Recommendation phase unless we have tests for all the normative 
langauge features, so please, members of TEST, do as Jeremy asks and 
generate these tests.
  -JH


At 1:10 AM +0100 9/24/02, Ian Horrocks wrote:
>On September 23, Jeremy Carroll writes:
>>
>>
>>  (owl:Nothing)
>>  Or Thing, or cardinality or one of the other OWL features.
>>
>>  One of my hopes with doing the Test form:
>>  http://sealpc09.cnuce.cnr.it:8080/wowg/jsp/edit.jsp
>>
>>  is that the members of the TEST focus area (myself, Dan Connolly, Jonathan
>>  Dale, Mike Dean, Libby Miller, Jos De Roo, Michael Sintek) and any other WG
>>  members, might find it easier to produce tests - particularly aiming to get
>>  coverage of OWL features. Most of them look straightforward e.g:
>>
>>  Nothing
>>  sameIndividualAs
>>  differentIndividualFrom
>>  samePropertyAs
>>  sameClassAs
>>  cardinality, minCardinality, maxCardinality
>>  hasValue
>>  someValuesFrom
>>  allValuesFrom
>>  unionOf
>>  intersectionOf
>>
>>  Producing one or tests for each of these would be good.
>>  It would be nice to have tests that illustrate the basic meaning 
>>and some easy
>>  misconceptions.
>
>I believe we also need some tougher tests, and in particular tests
>that show up subtle distinctions between entailment and
>non-entailment. I already provided a number of examples of such test
>(admittedly not in the relevant format).
>
>A nice example testing the interaction of functional properties with
>property subsumption is the famous "W" case. E.g., we have 5
>functional properties, with P1 subsuming both P3 and P4, and P2
>subsuming both P4 and P5. If an object x is related via P3 and P5 to y
>and z respectively, then (in the absence of other information) we can
>deduce that y and z are the same individual iff we know that x is also
>related to something via P4.
>
>By the way, it is not desirable for tests always to use individuals,
>as in many applications reasoning with classes will be enough (and
>reasoning with individuals is in general much harder, or at least much
>less tractable). The above example also works perfectly well for
>classes. I.e., in the absence of other information, the class of
>things that have (someValueFrom P3 C) and (someValueFrom P5 (not C))
>is NOT inconsistent, but the class of things that have (someValueFrom
>P3 C), (someValueFrom P5 (not C)) and (someValueFrom P4 Thing) IS
>inconsistent.
>
>Ian
>
>>
>>  e.g.
>>  Nothing
>>  <a> rdf:type owl:Nothing .
>>  is inconsistent.
>>
>>  e.g.
>>  the cardinality lot, crib something from FunctionalProperty tests.
>>
>>  e.g.
>>  hasValue a simple entailment showing that belonging to the 
>>restriction entails
>>  having the value
>>  a non-entailment shoing that belonging to the restriction and having some
>>  other value does not mean that the two values are equal.
>>
>>  e.g.
>>  unionOf
>>  Well we have to have the student/employee example! I guess we are 
>>coming round
>>  to trhat being an entailment rather than a non-entailment.
>>
>>  Tests that are received before the end of the week will get into the test
>>  document to be reviewed.
>>
>>
>>  Jeremy


-- 
Professor James Hendler				  hendler@cs.umd.edu
Director, Semantic Web and Agent Technologies	  301-405-2696
Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
Univ of Maryland, College Park, MD 20742	  240-731-3822 (Cell)
http://www.cs.umd.edu/users/hendler

Received on Monday, 23 September 2002 20:23:20 UTC