Re: New DQL Specification

Date: 06/11/02

  • Next message: Mike Dean: "Joint Committee telecon tomorrow 18 June"
    I would also like to support the point that we should not attempt to require 
    all answers implied by a kb.
    I would expect different reasoners to provide different reasoners and then 
    some reasoners that attempt to be more "useful" to also be able to "defend" 
    their answers.
    I would not specify the defense in our spec  but I expect the defense to 
    things such as the ability to explain the reasoning that the reasoner used to 
    generate an answer.  A defense might also include information concerning the 
    completeness of the reasoner and any attempt the reasoner may have made to 
    generate a type of answer - possibly the most specific answer(s).
    In a message dated 6/10/2002 11:25:06 AM Pacific Daylight Time, writes:
    > >All this brings up the basic point:
    > >
    > >     What is the definition of an answer?
    > >
    > >I don't think that this question has been satisfactorily answered.
    > Indeed, and that may be the basic problem we are having here. Your 
    > idea of answer apparently differs from Ian's, for example.
    > My own view on this is that there is no *definition* of an answer, 
    > but there are some conditions that all answers must satisfy. The only 
    > condition that everyone will likely agree on is that any answer must 
    > be entailed by the KB. (That is, that the instance of the query with 
    > that binding is entailed by the KB.)  I think in fact that this is 
    > *all* that should be required in a spec., and that any attempt to 
    > ensure that 'all' answers are given, or that logically equivalent KBs 
    > must give identical answers, are misguided in the SW context; they 
    > are far too strong to permit implementors to experiment, so will in 
    > fact simply be ignored in practice; and in any case I do not think 
    > that they are sensible in a Web open architecture. So I'm afraid that 
    > I find this debate somewhat pointless and timewasting.
    > Assuming that we do want to define something like the 'answer set' (I 
    > would like to see some rational explanation for why this concept is 
    > useful, by the way) , Ian has argued strongly that it should not 
    > include all 'answers' that can be logically inferred from the KB, but 
    > only those which arise from a binding of a query variable to a term 
    > in the KB Herbrand universe, in order to keep the inferential burden 
    > on the server within DL-manageable bounds. I am happy with that; but 
    > given the resulting incompleteness, it seems silly to object to a 
    > proposal on the grounds that logically equivalent KBs may not always 
    > deliver the same answers.
    > The point of the MID construction was to provide some useful 
    > information to the querying system in the case where a binding itself 
    > provides no useful information. We can debate the technical details 
    > of how best to do this, but that some such mechanism is needed seems 
    > to me to be obvious. In particular, any proposal for defining 
    > 'answer' which arbitrarily forbids a server from transmitting useful 
    > information to a querying engine, when it has already had to compute 
    > that information in order to answer the query, seems to me to be so 
    > mind-blowingly stupid as to not be worth discussing further.
    > Pat
    > -- 

    This archive was generated by hypermail 2.1.4 : 06/11/02 EDT