Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: SUO: Re: Conformance




Hi Matthew

I'll restrict this reply to the conformance stuff and do the rest 
some other time.

>  >
>>  As a general point, I think it is important that the SUO, like any
>>  standards effort, takes into account the likely needs of its intended
>>  user community. I have seen very little in the way of any discussions
>>  about who is likely to make use of the SUO, or any effort to contact
>>  any of them for their views. A standard that is used only by its
>>  authors may give them a feeling of authenticity, but that hardly
>>  seems sufficient to justify all the work involved.
>
>MW: There would certainly be value in defining some example use cases.
>I presume that we are trying to produce something that is "fit for
>purpose". Which rather presumes that you have a purpose in mind. Now
>I know the PAR says something about that, but it is hardly something
>you could measure any result of this activity against.

Why not? If that aim is considered hopeless, then I can see no 
apparent utility in having an IEEE/ISO standard at all. Why bother 
stating 'standards' that have no utility?

.....
>  > >>  > > >¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤~~~~~~~~~¤
>>  >>  > > >
>>  >>  > > > AP = Adam Pease
>>  >>  > > >
>>  >>  > > > AP:  - Implementations of SUO are "ontologies" or
>>  >>"information models".
>>  >>  > > >      - A conforming implementation is an ontology or
>>  >>information model that:
>>  >>  > > >        [#1] Uses terms as defined by the SUO, or
>>  >>  > > >        [#2] Uses terms that are defined (using SUO-KIF)
>>  >>  > > >             entirely by other terms in the SUO
>>  >>  > > >        [#3] Is consistent:  a contradiction cannot
>>  be derived by means
>>  >>  > > >             of  first-order logic from the set of
>>  statements belonging
>>  >>  > > >             to the implementation and the SUO.
>>  >  > > > >
>>
>>  There are several things that need to be corrected here. First, it
>>  obviously isn't appropriate to insist that conformance requires using
>>  ONLY terms from the SUO.
>
>MW: No, but it is only the use of the SUO, and not any additional material,
>about which conformance is relevant.

OK, that would be another way to define conformance, but that isn't 
what Adam seems to be saying above.

>  > After all, it is only supposed to be an
>>  UPPER ontology, not the whole thing. Conformance should require
>>  rather that all (?) terms 'fit underneath' terms from the SUO, maybe;
>>  but that would need to be made precise.
>
>MW: I don't think you can/should say anything about material people relate
>to SUO elements, only about the use of the SUO elements themselves. So if
>an ontology uses "class" it conforms to the SUO if and only if what is
>represented is precisely what is represented by "class" in the SUO.

How is that to be judged? That is, how does one determine that one 
representation has *precisely* the same meaning as another? That 
seems impossible to determine, in general. In any case, if my 
ontology uses a concept from the SUO and says something else about 
it, eg relates it to some other concept not in the SUO, then it is 
not at all clear that it is still *precisely* the same concept; at 
any rate, if it is, this judgement must rely on some theory of 
concepts which goes beyond the semantics of KIF.

>If
>the ontology has an object called "redBanana" we just don't care, and have
>nothing to say about it either to approve or disapprove.
>>
>>  Second, the notion of 'being defined' needs to be spelled out more
>>  exactly, particularly as (the last time I looked) SUO-KIF doesn't
>>  have any syntactic provision for stating definitions.
>
>MW: I brought this up. What I had in mind is concepts that are the
>intersection
>of sets, nothing more or less.

Well, then let us say that. That is much more restricted than the 
general notion of "being defined in terms of".

However, why are you now talking about sets? (How did *sets* get into 
the discussion?)

>So if in the SUO I had the sets "red" and
>"car"
>I could create a concept "redCar" that was precisely the intersection of
>those two concepts.

But not, say 'red or car', or 'car and not red' ? I'm willing to 
believe that is what you intend, but the motivation eludes me.

>  >
>>  Third, why the consistency condition? This seems to me to be quite
>>  inappropriate for a conformance clause. Part of the business of the
>>  conforming engine might be to detect inconsistencies, for example.
>>  (If inconsistencies are nonconforming, then that job is trivial,
>>  since there can't be any , by definition..) Also this places an
>>  uncomputable burden on any conformance tester; the problem even of
>>  detecting conformance isn't decideable.
>
>MW: I agree with you here.
>>
>>  Fourth, this entire approach to conformance seems to be only
>>  concerned with namespaces, and ignores the axioms and the logical
>>  language used.
>
>MW: No, it is the axioms that define what the concepts mean, or at
>least what rules/constraints need to be satisfied. You cannot separate
>the strings from the axioms in which the strings occur.

Well, I tend to agree, in abstract terms; but again, Adam's 
definition makes no reference to axioms. Maybe it should. However, 
nothing in KIF says anything about the relationship between axioms 
and rules/constraints. There is no KIF syntax for even stating rules 
or constraints.

>
>>  Suppose someone writes a block of DAML and uses the
>>  SUO namespace: does that constitute conformance?
>
>MW: It could do, but it is not sufficient.

How could it, if conformance refers to axioms, and they are written 
in a different language? The question of whether or not a set of 
mixed KIF and DAML assertion is or is not consistent isn't even a 
well-defined question to ask; it is meaningless.

There seem to be all kinds of implicit assumptions in these 
conformance ideas that have not been spelled out. Until they are 
spelled out, we simply do not know what we are talking about.

>
>>  Or must what they
>>  say about the named concepts also conform to the SUO axioms?
>
>MW: Precisely.

But how could we possibly tell? DAML and KIF have different model 
theories. DAML wouldn't even be parsed by a KIF engine: it is 
KIF-gibberish.

>
>>  How is
>>  *that* kind of conformance specified?
>
>MW: Well if they were using the SUO as an index, they could
>just state that the concepts do conform, and
>just refer to the standard. Or they might code the constraints
>in JAVA for some other purpose.

How could one judge that a piece of JAVA code was in conformance to 
some KIF axioms?

>  > Must they use SUO-KIF?
>
>MW: They must be able to demonstrate that their implementation
>behaves in the way that the SUO-KIF statements specify.

But SUO-KIF does not specify ANY behavior. It is an assertional 
logic, not a programming language or process description language.

None of this discussion makes any sense at all. Lets get concrete and 
then maybe we can make some progress. Suppose I write some code and I 
want to know if it conforms to the SUO standard. How exactly do I set 
about trying to find out? Or, if I claim that it does, how would 
someone prove me wrong? What kind of thing would constitute 
nonconformance? Don't answer in 'ontological' terms, ie by talking 
about 'using a concept'; that isn't well-defined enough, since there 
is no way to even say what it means, in general, for a program to be 
'using' a 'concept'.

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