- From: Dave Reynolds <der@hplb.hpl.hp.com>
- Date: Mon, 12 Sep 2005 13:35:38 +0100
- To: Giorgos Stamou <gstam@softlab.ntua.gr>
- CC: "'Sandro Hawke'" <sandro@w3.org>, public-rule-workshop-discuss@w3.org
Giorgos Stamou wrote: >> [der] >>Is this a question we could pose to the Semantic Web Best Practices >>working group? > > > [Giorgos Stamou] > > Well, I think this is a good idea. And since I am a member of the Semantic > Web Best Practices Working Group (also co-chairing a new Task Force on > Multimedia Annotation in the Semantic Web, an area that needs uncertainty > representation) I could certainly start this discussion there, as well. That sounds good, thank you. I'll look forward to seeing how the discussion goes. I think the issue of how to handle uncertainty on the semantic web is a bigger question than how to generalize logical rules. > The work on fuzzy logic has been started in 1965. And, in its late form (see > the book [Petr Hayek, Metamathematics of Fuzzy Logic, Kluwer Academic > Publishers 1998]), it is really a general framework covering several other > relevant theories of multi-valued logic. It is a clear and sound framework. I last used it in the early 80's so I'm aware it has been around for a while. > Still, it is not complete! But I think most theories in logics are not > "complete". I also think that the W3C Standardization process does not only > cover "complete" theories. Moreover, there is no final "agreement" of the > community that discusses the charter, in basic issues of classical logic > (consider the example of the usefulness of monotonic or non-monotonic > logic). > So, I understand your argument but I don't really think that we are very > confident with several other language extensions fairly included in the > charter. Moreover, obviously I did not propose uncertainty for the core > language. I only proposed to be included in the strongly-motivated in-scope > extensions to be considered. Perhaps you are right that uncertainty representation is as settled an issue as how to combine non-monotonic rules with RDF and OWL but the latter is hardly an uncontroversial issue :-) >> [der] >>For good or ill, the emphasis of the draft charter is not on an >>implementable language but on an interlingua for exchange of rules >>between existing systems. It's not clear that the uncertainty use cases >>really fall into the interlingua remit. I would have thought that the >>issues of successful exchange and integration of rules involving >>uncertainty reasoning would be much greater that those of simply using >>fuzzy inference within a single system. Whereas the use cases seemed to >>want an implemented language they could use without a particular need >>for rule exchange. > > > [Giorgos Stamou] > I think they fall in the same framework. In the case you have different rule > systems dealing with uncertain and fuzzy rules, they should exchange these > rules in a similar form, in order to be able to deliver the uncertainty and > fuzziness. The point I was trying to make was that for non-fuzzy rules there are a lot of languages in use which people seem to want to exchange rules between, so I can believe that it makes sense to develop an interlingua now. Whereas it's not so clear that there are so many fuzzy rule languages that fuzzy-rule exchange is currently the presssing issue. It seemed more like the uncertainty use cases were asking for *a* solution rather than a way for their many solutions to interoperate. >>[der] >>The stated requirement to "generalize the two-valued Boolean logic of >>{0,1} into the interval [0,1]" seemed somewhat directed towards a fuzzy >>logic solution. > > > [Giorgos Stamou] > If it is because of the "[0,1]" requirement we could change it, although > fuzzy logic will also follow it (other relevant theories would not). > However, I think "[0,1]" is more preferable since it is a reasonable > extension of {0,1}. The "[0,1]" bit is part of it. In addition, the use cases presented seemed like they might be amenable to statistical techniques and were in domains where, IMHO, a precise understanding of the probability of an incorrect recognition is mandatory. It's not clear that all such statistical procedures can be thought of as generalizations of boolean logic. > I understand that some language extensions will not be included and I > respect your opinion that there is no clear need for the uncertainty > extension to be included from the beginning (although if I correctly > understand you are positive to be included in the future). I also recognize > the very difficult and critical work of people writing the charter in order > to fairly balance it between several other people needs and wants. > But let me insist that it is fair enough to include the issue of > uncertainty, which was mentioned a lot by industrial participants in the > Workshop (which was organized mainly in order to find the industry > requirements), nobody is really against it, and does not change the work of > anybody. Fair enough, I think I've expressed my reservations sufficiently and will shut up now. Dave
Received on Monday, 12 September 2005 12:36:29 UTC