- From: Jim Hendler <hendler@cs.rpi.edu>
- Date: Mon, 18 Aug 2008 09:50:02 -0400
- To: Ian Horrocks <ian.horrocks@comlab.ox.ac.uk>
- Cc: public-owl-wg Group WG <public-owl-wg@w3.org>
I'm still confused - how are we to differentiate between intentional triple that take us out of DL and mistakes? According to the below, any time I do anything that takes an ontology out of DL, a conformant OWL tool is welcome to decide I want to be in Full only. We invite tool vendors to do something (but t cannot be required) - just to put some numbers against this, when we did the ontology study at Md a couple of years ago, the majority of OWL DL ontologies were actually not in DL until fixed. The most common error was that someone would refer to an external triple but not provide a "type" for it. There were also a number of ontologies that did metamodeling (some, but not all of which would now be handled by punning if I understand right), where the user would want to be in Full (since it was disallowed in DL) -- I understand why the WG is against some method of signaling intentions in general, but for this one particular case, why not do something unambiguous and easier for humans to remember and machines to recognize? Remember, in the outside world a lot of OWL users are not perfect and do make mistakes - having those mistakes put them in Full instead of DL seems odd to me (and note that a tool may not be interacting with the original developer or user -- so if I am using a tool, pull some ontology from a library, and my system says "do you want this in DL or Full?" how do I know the answer? Chances are I wold look in the comments to see if the person said -- so again, why not make it easier by having something explcit? -JH On Aug 18, 2008, at 7:27 AM, Ian Horrocks wrote: > > What was being suggested was only to offer the advice/observation > that including a triple that pushes an ontology outside DL would be > one way for a user to signal their intention that the ontology be > interpreted using Full semantics even if it is otherwise > syntactically DL. Of course tools would still be free to ignore this > and apply "repairs". > > Given that we are only proposing to offer advice/observation, I > don't think that we should insist on a specific triple -- I would > worry that this would be interpreted as saying that tools should/ > must check for this triple and behave accordingly. Better, I think, > to stick with the same kind of informal advice but to augment it > with the observation that it is advisable to use a triple such as > sameAs sameAs sameAs that is unlikely to arise by accident or be > "repaired" by tools. Given that we all agree that what we are > discussing here is very much a corner case, I hope that this will be > acceptable to all concerned and that we don't need to spend more > time on this issue. > > So, in the light of our discussion, the revised proposal to resolve > would be as follows: > > As per discussions and strawpolls at the Boston F2F [1], the 6th > August telecon [2], and in email [3], I propose that we close this > issue by adding to the spec the following advice/observation: > > Note that users wanting to signal their intention that an ontology > be interpreted by tools using the OWL RDF semantics can do so by > adding a triple that takes the ontology out of OWL DL but does not > otherwise affect its meaning. It is advisable to use a triple that > is unlikely to occur unintentionally, and so unlikely to be > interpreted as an error and "repaired" by tools. An example of such > a triple is sameAs sameAs sameAs . > > Regards, > Ian > > [1] http://www.w3.org/2007/OWL/meeting/2008-07-28#Strawpoll_on_signaling_semantics > [2] http://www.w3.org/2007/OWL/meeting/2008-08-06#Strawpoll_on_resolving_issue__2d_111 > [3] http://lists.w3.org/Archives/Public/public-owl-wg/2008Aug/0049.html > and thread > > > On 12 Aug 2008, at 13:10, Jim Hendler wrote: > >> OK, so supposing I'm an OWL DL user and I want to use a DL >> reasoner. I accidently, however, assert something that puts the >> ontology in Full (and when we did the analysis of 1500 ontologies 3 >> years ago, there were at least 100 for which this case was true -- >> usually because someone referred to something from a remote name >> space without adding the appropriate type or imported something >> that put them into Full without their realizing it) -- so according >> to this, tools like Pellet, instead of "fixing" these mistakes >> (heuristically) would now need to assume the person knew what they >> were doing and that they want to be in Full -- so it would be rare >> that they want to do this on purpose, but not rare that it would >> happen by accident. >> My point is not that I think there shoudln't be some way to do >> this, but rather that it should be explicit -- otherwise we get in >> a situation where tools will assume it's a mistake, and then the >> user will have to do something extra to make it clear they meant it >> So my argument is not that we should have some way to always signal >> intended use, but that for this corner case, we should have >> something unambiguous that is not likely to be used by mistake >> (which is why, sameAs sameAs sameAs seems appealing) -- but we >> should have one specific way to do it or else we create the >> potential for use confusion and non-interoperability since we'll >> end up with computers trying to guess human intentions -- something >> they are not good at >> Seems to me the easy solution in this case is to be insistent that >> some very specific triple get asserted in this rare case (and not >> that we have a general mechanism for intended use) which would >> solve Bijan's problem of creating something overly general (and, >> since it would only be used in this case, the complexity of >> "conflicting" intendedUse Triples is a non-issue) >> -JH >> >> >> >> On Aug 11, 2008, at 2:50 PM, Ian Horrocks wrote: >> >>> >>> The proposal *is* that there would be no specified way to signal >>> the intended semantics, so tools would have their choice. Thus I'm >>> not sure why you don't like it. Perhaps I didn't explain it very >>> well (in my defence I should say that I was only trying for a very >>> quick summary of the cited discussions). >>> >>> The idea is to include only the advice/observation that including >>> a triple that takes the ontology out of OWL DL obviously forces it >>> to be interpreted using the RDF semantics, with "sameAs sameAs >>> sameAs" being given as an example of such a triple (I should have >>> said "for example" instead of "namely"). >>> >>> Regards, >>> Ian >>> >>> >>> On 8 Aug 2008, at 17:33, Jim Hendler wrote: >>> >>>> Having left the WG, I seem to still be being pulled in a lot by >>>> side emails, so let me state, as RPI AC rep, that we don't like >>>> this solution. I see two problems >>>> 1 - it seems to us that people who use DL are more likely to >>>> understand the difference between DL and Full than those who are >>>> just using the vocabulary, so the chances of this triple being >>>> included seem very low - thus, we'd prefer to see someone who >>>> understands that they want to be only DL should have to do >>>> something to signal that >>>> 2 - by this decision, if a user accidently does something to make >>>> their ontology OWL Full, they will be signaling they only want to >>>> be in Full (since it says "should include a triple that takes the >>>> ontology out of OWL DL") -- if the meaning is that we want users >>>> to use only this specific triple, then it seems to me we should >>>> do something more obvious, like putting in some semantics free >>>> definition that expresses intent -- i.e. instead of "sameAs >>>> sameAs sameAs" wouldn't it be a lot smarter for the document to >>>> include "[] intendedUse OWL-Full"? >>>> >>>> In fact, given these two factors, it seems like we should either >>>> have explicit means for signaling semantics when intended, or no >>>> specified way, meaning tools have their choice. >>>> -JH >>>> AC Rep, RPI >>>> >>>> >>>> On Aug 8, 2008, at 11:31 AM, Ian Horrocks wrote: >>>> >>>>> >>>>> As per discussions and strawpolls at the Boston F2F [1] and our >>>>> most recent telecon [2], I propose that we close this issue by >>>>> adding to the spec the advice that users wanting to ensure that >>>>> their ontology is interpreted *only* as OWL Full should include >>>>> a triple that takes the ontology out of OWL DL, namely: >>>>> >>>>> sameAs sameAs sameAs . >>>>> >>>>> Regards, >>>>> Ian >>>>> >>>>> [1] http://www.w3.org/2007/OWL/meeting/2008-07-28#Strawpoll_on_signaling_semantics >>>>> [2] http://www.w3.org/2007/OWL/meeting/2008-08-06#Strawpoll_on_resolving_issue__2d_111 >>>>> >>>> >>>> "If we knew what we were doing, it wouldn't be called research, >>>> would it?." - Albert Einstein >>>> >>>> Prof James Hendler http://www.cs.rpi.edu/~hendler >>>> Tetherless World Constellation Chair >>>> Computer Science Dept >>>> Rensselaer Polytechnic Institute, Troy NY 12180 >>>> >>>> >>>> >>>> >>> >>> >> >> "If we knew what we were doing, it wouldn't be called research, >> would it?." - Albert Einstein >> >> Prof James Hendler http://www.cs.rpi.edu/~hendler >> Tetherless World Constellation Chair >> Computer Science Dept >> Rensselaer Polytechnic Institute, Troy NY 12180 >> >> >> >> > > "If we knew what we were doing, it wouldn't be called research, would it?." - Albert Einstein Prof James Hendler http://www.cs.rpi.edu/~hendler Tetherless World Constellation Chair Computer Science Dept Rensselaer Polytechnic Institute, Troy NY 12180
Received on Monday, 18 August 2008 13:50:45 UTC