- From: Jeff Heflin <heflin@cse.lehigh.edu>
- Date: Wed, 08 May 2002 14:19:10 -0400
- To: Pat Hayes <phayes@ai.uwf.edu>
- CC: www-rdf-logic@w3.org
While I agree that the idea of saying what not to import is an interesting one, I don't think it is a scalable solution. Every time somebody comes up with a incorrect ontology, everyone who is concerned about making sure their documents are understood correctly will have to add a line to forbid importing it. I can envision a hacker that writes a script which creates thousands of such ontologies every second. Imagine the problems that would ensue. As for the agent semantic negotiation idea, I am skeptical. It may be possible, but I don't see it happening anytime soon. It sounds like a really tough problem. Now, others may be equally skeptical of my imports solution. I'll admit it may not be perfect, but I think it is a good starting point, and I've already implemented a prototype that handles the basic ideas. Jeff Pat Hayes wrote: > > I agree, that is the only sensible way to look at it. If/when we get > around to putting something like this into the CL standard it will > definitely be a special syntax, not a normal logical assertion. If it > were an assertion it would have to be in a special meta-level > concerned with ontologies as entities, and I don't think anyone wants > to get involved with that. > > However, I still wonder if we need it. Consider the admittedly naive > assumption that anyone who uses a vocabulary is committed to using it > the same way. Then we don't need importing: just use the same names > is all you have to do. OK, so it is going to break down at times; but > I bet 'imports' is going to break down at times as well, because > people will misunderstand the intended meanings in complex > ontologies, and so on. Seems to me that the only real purpose of > having an 'imports' tag is to be able to NOT import some stuff you > DONT want to agree to. So it might be more use, in fact, to have that > as the primitive. That is, instead of having as a default that you > might not agree with anything that you don't explicitly mention as > imported, the SW might actually do better to base itself on the > riskier, but more interesting, presumption that you agree with > anything that you don't explicitly disagree with. Then engines could > just troll around looking for content and put it together to draw > conclusions, and could probably get a lot further than if they are > restricted to the narrow channels of explicit import links. It will > go wrong sometimes, but we have far harder interoperability problems > to solve (involving genuine conceptual mismatchings), so I think a > few simple disagreements are the least of our worries. > > Q: What if my engine finds a contradiction between stuff it got from > ontology A and ontology B? A: Well, it could just ignore them; or it > could try to decide which of them was more likely to be right (using > dates, maybe, or a Googlish kind of how-many-others-agree criterion) > and maybe send them a notice telling them about the problem, if there > was some way to do that. Then one of them might want to change, or > to exclude the other. > > This suggests a vision of an SW which is in a constant process of > self-repair. Imagine a kind of truth-maintenance protocol that > allowed agents to negotiate differences of opinion. With a bit of > care and a bit of luck, the SW might get to be more reliable in its > opinions than the people who wrote the ontologies in the first place. > At any rate, it could sort out sub-communities of ontologies which > all use vocabulary terms in a consistent way. This would also, it > occurs to me, provide some security against malicious bad-data > ontologies which someone will almost certainly think of creating once > the SW gets real. > > Pat Hayes > > -- > --------------------------------------------------------------------- > IHMC (850)434 8903 home > 40 South Alcaniz St. (850)202 4416 office > Pensacola, FL 32501 (850)202 4440 fax > phayes@ai.uwf.edu > http://www.coginst.uwf.edu/~phayes
Received on Wednesday, 8 May 2002 14:19:18 UTC