- From: Ulrike Sattler <sattler@cs.man.ac.uk>
- Date: Fri, 9 Mar 2007 15:03:07 +0000
- To: Alex Tucker <alex@floop.org.uk>
- Cc: public-owl-dev@w3.org
On 8 Mar 2007, at 16:53, Alex Tucker wrote: > > Alan, >> What i have done, for non-compound keys, is create a uri with the >> value embedded as a string, use and object property and make that >> inverse functional. > We've been following a similar pattern, essentially preferring > ObjectProperties over DatatypeProperties for the most part. However, > Ian's assertion that the "tractability of reasoning ... depends on the > fact that they are typically *not* used in this way," worries me a > little, if, as users, we've all been expecting IFPs to cope. > let me explain this dependency a little: IFPs (whether they are present explicitly or via the work-around described by Alan) should be harmless - in case where you have individual names (even many of them), say a1, a2, ...., a1000000,... and all of the are related via the datatype property "hasID" to some integer, and you have declared hasID as inverse functional: now, in case that there are 2 individuals, say a17 and a23, that have the same hasID-filler, then a17 and a23 will be identified. and they can be dangerous (wrt current technology and to the best of my knowledge and only unless somebody comes up with some clever optimisations) if - we have "many" existential restrictions, e.g., when (describing infinite lists) we say that every listelement has a next-successor who is a listelement, and that next is inverse-functional, and that every listelement has an id-successor who is a non-negative integer smaller than 1,000,000...now if we would also say that id is inverse- functional, then this would mean that we could have 1,000,000 listelements, but not 1,000,001, and thus we would either have to generate all 1,000,001 listelements, then guess which of them are the same to try to satisfy all our constraints, then fail -- or we would need to work with such "global cardinality" constraints (atmost 1,000,000 listelements) directly - and our current reasoning techniques can't do the latter and will fail trying to do the former. - now, if you consider the previous example, we could make it more complicated by saying that "odd" listelements have "even" listelements as next-successors, and that "even" listelements have odd listelements as successors, ... and then later say something about the id-successors of odd things, e.g., that they are below 1,000,000: now it is even more tricky to see how many listelements we can have! Cheers, Uli > Alex. > >
Received on Friday, 9 March 2007 15:06:16 UTC