- From: Carsten Lutz <clu@tcs.inf.tu-dresden.de>
- Date: Thu, 13 Dec 2007 22:12:55 +0100 (CET)
- To: Uli Sattler <sattler@cs.man.ac.uk>
- Cc: Alan Ruttenberg <alanruttenberg@gmail.com>, "Peter F. Patel-Schneider" <pfps@research.bell-labs.com>, public-owl-wg@w3.org
On Thu, 13 Dec 2007, Uli Sattler wrote: > > On 13 Dec 2007, at 16:04, Alan Ruttenberg wrote: > >> >> I have had a discussion with Uli about this, and it seems that a limited >> form of this would not compromise decidability - e.g. no binary comparisons >> other than equality. She said she would write up a proposal. > > ...I seem to remember that i agreed to have a think (!) about it - and I had: > it seems to me that (i) we would need to "fork" OWL11 in a difficult to > understand way (i.e., you can either use datatype properties at the end of > property chains or, say, comparisons, but not both), and (ii) we would need > to ask implementors whether they would be willing to implement this. I would > thus favour to not allowing them at all....Carsten? I basically agree with your assessment. Adding datatype properties at the end of property chains (lets call this "datatype chains") is very close to what is called "full concrete domain support" in http://lists.w3.org/Archives/Public/public-owl-wg/2007Nov/0197.html but with equality as the only binary comparison. Even if we have both datatype chains and binary comparisons, reasoning will stay decidable. However, this combination poses serious challenges to the reasoner developers. I think we better avoid it. Without binary comparisons, I assume that datatype chains are ok from an implementation perspective. On the other hand, I believe that forking OWL DL should be avoided at all costs. BTW, I disagree with Alan that there are not many use cases for comparisons. For example, Sebastian gave many (and convincing ones) at the F2F. Uli, didn't he also send a mail with some use cases? greetings, Carsten > Cheers, Uli > >> -Alan >> >> On Dec 13, 2007, at 9:56 AM, Peter F. Patel-Schneider wrote: >> >>> >>> Issue-8 asks for property chains that end with data properties. >>> >>> Adding this construct to OWL 1.1 would compromise decidability. >>> This feature would automatically be in an OWL Full version because in >>> OWL Full data properties are also object properties. >>> >>> Later discussion asked whether having data properties in the middle of >>> a chain can be done. >>> >>> In OWL 1.1 such chains would have an empty extension, and thus be >>> useless. The situation in OWL Full is the same as for data properties >>> at the end of a chain. >>> >>> I therefore propose that we CLOSE ISSUE-8 (even though it is not even >>> OPEN) without doing anything on the twin grounds that it both >>> compromises decidability in OWL 1.1 and is not handled by tools, and >>> that there is nothing special that needs to be done in OWL Full. >>> >>> Peter F. Patel-Schneider >>> Bell Labs Research >>> >>> PS: I'm proposing handling ISSUE-8 in this manner as is it is closely >>> related to ISSUE-83. >>> >> >> > > -- * Carsten Lutz, Institut f"ur Theoretische Informatik, TU Dresden * * Office phone:++49 351 46339171 mailto:lutz@tcs.inf.tu-dresden.de *
Received on Thursday, 13 December 2007 21:13:15 UTC