- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Fri, 18 Apr 2008 15:27:22 +0100
- To: John Goodwin <John.Goodwin@ordnancesurvey.co.uk>
- CC: Kendall Grant Clark <kendall@clarkparsia.com>, public-owl-dev <public-owl-dev@w3.org>
Disclaimer, all I know about RCC8 I gleaned in thirty seconds from http://en.wikipedia.org/wiki/Region_Connection_Calculus There are two very different problems here. one is whether given an ontology using such RCC8 primitives one can implement the composition table, which seems to have a lot of disjunctive terms over properties, which, at first glance appears to be byeond the expressivity of OWL2. A second problem is whether given regions expressed geometrically (rather than topologically), for example, by a set of coordinates, or as a definition of an oval (a centre and eccentricity), or whatever, can you compute such RCC8 relationships, and given class definitions that depend on these can you verify ontological relationships. I would be deeply surprised if this were possible. One would expect useful regions to be defined by equations that were non-linear, so that resolving class expressions using mixed tableau and numeric programming would involve non-linear programming, which is not sufficiently well behaved to get predictable behaviour from an OWL reasoner. If one restricts region definition to polygons (with straight edges), with rational coordinates then the numeric part would be linear programming that would be tractable in theory. But the infinitissimal difference between TPP and NTPP for example, would be lost in rounding errors, and, also impact the performance (LP is polynomial in the size of the program definition - as one increases the precision of the coordinates then the rationals used to express the program require more and more bits to express them, and the LP algorithm takes longer and longer, despite essentially being used on the same problem). But then I am the n-ary datatypes pessimist :) Jeremy
Received on Friday, 18 April 2008 14:29:09 UTC