- From: Jim Hendler <hendler@cs.umd.edu>
- Date: Wed, 22 Jun 2005 18:04:50 -0400
- To: Michael Kifer <kifer@cs.sunysb.edu>, Bijan Parsia <bparsia@isr.umd.edu>
- Cc: public-sws-ig@w3.org, dreer@fh-furtwangen.de, Jos de Bruijn <jos.debruijn@deri.org>
My reply to this thread is now in rdf-rules -JH At 15:49 -0400 6/22/05, Michael Kifer wrote: >Bijan Parsia wrote: >> >> Perhaps this should move to rdf-rules? > >Perhaps. > > >> On Jun 22, 2005, at 11:29 AM, Michael Kifer wrote: >> >> > Bijan Parsia wrote: >> [snip] >> >> To extend the conversation in another direction, is there any reason >> >> to think >> >> that a logic programming paradigm, in general, is the right approach >> >> to nonmon >> >> on the Web? Representationally? There are many non-monontonic >> >> formalisms >> >> (consider default logic and autoepistemic logic) and it might be that >> >> they 1) are >> >> better for web contexts and 2) play better with owl. (It's plausible, >> >> for example, >> >> to think that default logic can be made to fit better because of the >> >> separatation >> >> of the base representation and the default rules. Even there, >> >> adjustements must >> >> be made.) >> >> >> >> (Of course, anything in this space runs into the problem that, in >> >> general, >> >> nonmon formalism are much more computationally difficult than >> >> corresponding >> >> monotonic ones. The LP position often appeals to the >> >> scalablility/computational >> >> goodness of, say, deductive databases. But if that comes at the price >> >> of >> >> throttling back expressivity forever...maybe it's not such a great >> >> idea. Pat Hayes >> >> often, to my understand this, as thinking of nonmon constructs as >> >> part of the >> >> *data* on the web (to his mind, bad), and nonmon as a way of >> >> *reasoning with* >> >> the data on the web (good...it's located in the agent or processor >> >> which is in a >> >> position to make certain assumptions with a good sense of the risks)). >> > >> > These are all valid points for future research. >> >> That's the extension of the conversation I'm after. > > >I think it is a research program. I don't think much, if anything, can be >accomplished on a mailing list. > > >> > I believe, however, that >> > >> > 1. It is naive to assume that one single formalism like DL or LP would >> > serve the humankind forever. >> >> I certainly don't think that. > >Good! > > >> > The architecture should provide for multiple formalisms (where the >> > formalism would be identified together with the statements -- RuleML >> > attempts to do something like that). The communicating parties will >> > either be able to talk (if they both understand that particular >> > formalism) >> > or they won't, but at least they will know it. >> > Certain degree of interoperability between the different formalisms >> > can >> > be provided without them being built on top of each other. >> >> This is going in a different direction, which I'm sympathetic too. But >> it seems to end up in the land of multiagent systems (with agents >> wrapping and mediating different data sources). That *doesn't* seem >> like the semantic web as I've heard it articulated. >> >> Maybe the semantic web as such is impossible! > > >I don't know what "semantic web as such" is, but I do believe that >defining it as a single stack is doomed to fail. > > >> > 2. Regarding the suitability of LP, this is backed by over 30 years of >> > practice. >> >> Hmm. Yes and no, right? The question is suitability *for what*. Of >> course, we're all groping in the dark, really. > >LP has been used for knowledge-intensive apps for as long as I stated. >Webby things are not that different -- just another application. There are >interesting problems for sure, but it is funny to watch some of the >discussions in which people create fetishes and pray to the god of URLs. > >> The W3C made a bet >> though that is not easily reconcilable with LP (and components of >> which, at least, have similar depth in background). So, do we zig? Zag? >> Stay the course? Stay mostly the course? Start over? > >So, they erected one stack - this is fine. If they will insist that this is >the one and only true stack, then the thing is going to die due to >irrelevance. > > >> > Default logic is nice, but it is just a theoretical tool at >> > this point. Before it (or its derivatives) can make into a Web >> > standard, >> > I suggest to give it a try (or **practical** use) for, say, 10 >> > years by a >> > reasonably sized user community. >> >> While that would be my general suggestion for *EVERYTHING* :), betting >> seems to be the name of the game. > >See, organizations like OASIS let their standards to die. If W3C is fine >with that then they can bet all they want. >But it seems to me that W3C is not prepared to kill its own standards, and >in this case it should bet very carefully and with an eye on the future. > >> >> Looking at LP land, I don't see systems doing the "Web" thing. Of >> course, I'm not entirely sure what the web thing *is* really. I'd love >> to have better clarity on that so we could figure out what really >> *should* be going on. > >Exactly. What is the "Web" thing precisely? >If you view it as a large distributed KB then LP is arguably doing this. > > >> However, and I think it's a reasonable position, you are actually >> advocated a non-integration strategy. (As you said in one.) That's >> fine, but then I would like it if those cards were laid on the table >> instead of claims of integration, overlap, compatibility (let me note >> that you are not the one making such claims). Let's change the freaking >> architecture to a hub and spoke, or whatever. > >As I said, it is presumptuous to claim that the current technology will >remain true for all times -- even for 5 years from now. A realistic >architecture should allow for more spokes. > >> Why two stacks instead of >> twenty and how do you make those twenty talk *at all*? > >One way for them to talk is to allow them to view each other as black boxes >and send queries to each other. This is essentially the architecture of >AL-Log. The Eiter et all. papers that you cite in your paper take the same >approach and try to integrate DL with LP a bit tighter. > >It is strange that you even cite Eiter's paper because, if anything, this >paper is an argument that a single stack is a bad idea and that several, >loosely integrated, stacks is a way to go. > > >> This is close to the RuleML view of things. I've watched RuleML for >> quite a while and I still believe that its approach, while appeal, is >> not the kind of thing that the W3C likes to do. They like to pick >> winners, rather than pick integration formats. (Of course, they like >> picking winners at the "right" level...XML is pitched, after all, as an >> integration format!) > >Winners? How do they determine who the winner is? (The rest of the diatribe >is not for a public list :-) > > > --michael -- Professor James Hendler Director Joint Institute for Knowledge Discovery 301-405-2696 UMIACS, Univ of Maryland 301-314-9734 (Fax) College Park, MD 20742 http://www.cs.umd.edu/users/hendler
Received on Wednesday, 22 June 2005 22:05:41 UTC