Re: Possible semantic bugs concerning domain and range

>There may be pragmatic/implementation reasons to go for implies
>semantics in all cases:
>
>- it can only lighten the burden on implementors as there will be
>fewer kinds of logical entailment to worry about.
>
>- the cost isn't very great as implied functionality, transitivity
>etc. due to strange constraints on possible models doesn't seem like
>it would be of great interest.
>
>- it would satisfy Pat's complaint that logically entailed range and
>domain restrictions are positively harmful.

But it would fail to satisfy the RDFS requirement that subClassOf and 
subPropertyOf are transitive. We could of course just add this as an 
ad-hoc semantic requirement, but that seems very tacky.

Given the option between all IF and all IFF, I think the all-IFF 
option is more coherent. But I would prefer a more tailored solution, 
as you know.

Pat



>Ian
>
>
>On October 15, Jeremy Carroll writes:
>>
>>  Summary: attempt to collect arguments about this issue.
>>  (Also added justification for uniformity, and a new argument about mutually
>>  entailing ontologies).
>>
>>  >Range
>>  >Domain(P,C) implies/iff (forall x,y P(x,y) -> C(x))
>>
>>  >TransitiveProperty(P) implies/iff (forall x,y,z (P(x,y) ^ P(y,z)) 
>>-> P(x,z))
>>  >SymmetricProperty(P) implies/iff (forall x,y P(x,y) -> P(y,x))
>>  >FunctionalProperty(P) implies/iff (forall x,y,z (P(x,y) ^ P(x,z)) -> y=z)
>>  >InverseFunctionalProperty(P) implies/iff (forall x,y,z (P(y,x) ^
>>  >P(z,x)) -> y=z)
>>  >inverseOf(P,Q) implies/iff (forall x,y P(x,y) -> Q(y,x))
>>
>>  I hear Dan, Jos, myself, Peter and Ian being able to go either way here.
>>
>>  There seem to be various arguments:
>>
>>  - treat them all the same
>>  (unarticulated)
>>   Less difficult for implementors,. less difficult to document, 
>>less difficult
>>  to learn. I suspect the Guide would be shorter with iff semantics.
>>
>>  - implies only
>>   Few implementation would actually implement iff.
>>   (However most of the implementors in the group seem to have come 
>>round to the
>>  possibility of implementing iff)
>>
>>  - natural usage
>>   Pat (so far unsupported) has opinions about natural usage that 
>>split domain,
>>  range and inverse off as intensional (implies) and the others as extensional
>>  (iff).
>>
>>  - rdf datatyping
>>   I think this argument is now dead - some versions of 
>>rdf:datatyping requried
>>  intensional reading of rdf:range.
>>
>>  - possibility of identifying identical ontologies (new argument)
>>   With extensional semantics then ontologies using these with identical
>>  semantics entail one another. With intensional semantics then it is not the
>>  case e.g.
>>
>>  <owl:FunctionalPropery rdf:ID="a">
>>     <owl:inverse rdf:resource="#b" />
>>  </owl:FunctionalProperty>
>>
>  >
>  > <owl:InverseFunctionalPropery rdf:ID="b">
>  >    <owl:inverse rdf:resource="#a" />
>  > </owl:InverseFunctionalProperty>
>  >
>  > either have identical meaning or not.
>  > Seems potentially useful, to say that they do have identical meaning.
>>
>>  - argument by authority
>>  iff we take this style of argument seriously
>>
>>  - surprising entailments
>>  An empty property is necessarily transitive, functional, inversefunctional,
>>  its own inverse,  etc.
>>
>>
>>  I think consistency is what I feel strongly about.
>>
>>  Jeremy
>>
>>
>>
>>
>>
>>
>>


-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes

Received on Thursday, 24 October 2002 12:20:25 UTC