Re: More results

On September 15, Sean Bechhofer writes:
> Sandro -
> Some more test results data from tests that Ian tried with FaCT is
> available at [1]. Note that as FaCT lacks some of the reasoning services
> required to pass the tests (such as an ABox and datatype support), and is
> instead is given a (manual created but still) equivalent problem, these
> shouldn't be taken as real "passes". However, the results may be
> informative.

I think that Sean's comments about these results are a bit too
negative. The results in question were obtained using a sound and
complete mapping from a subset of OWL DL/Lite to a FaCT Tbox - i.e.,
for those tests that I was able to translate, the answers should be
both sound and complete (modulo theoretical or implementational bugs
in FaCT).

Some of the testing was done by hand, e.g., I used Sean's validator to
map OWL to DIG (an XML syntax used by several DL reasoners), pasted
the result into FaCT, and typed the answer into a text file. It would
have been trivial for a competent hacker to automate this process.

One interesting aspect of the translation is the treatment of datatype
properties. Jeremy has transformed many of the tests into OWL Lite
using a mapping similar to the one I suggested some time ago (can't be
bothered hunting up the relevant email). For reasons best known to
himself, he chose to use datatype properties to enforce disjointness
(e.g., he makes A disjoint from B by asserting conflicting cardinality
constraints w.r.t. a datatype property). FaCT doesn't support datatype
properties. However, given this very limited use of datatype
properties, they can simply be treated as object properties without
compromising soundness and completeness. Tests that involve more
extensive use of datatype properties have not been attempted using

Regards, Ian

> 	Sean
> [1]
> -- 
> Sean Bechhofer

Received on Monday, 15 September 2003 17:29:59 UTC