Re: [OEP] comments on n-ary relations document

Pat

Pat Hayes wrote:

>> Pat
>
>> Thanks.  Clear and short and makes the point that the key difference is that
>> representation involves sequence.  Minor issue.  Both the properties
>> "rest_of_list" and "first_item" need to be functional to avoid including the
>> possibility of trees in the pattern.
>
>  Not sure I agree. This would make sense if one were to think of it as a kind
> of syntactic check, but it would not have that effect in OWL; and if someone
> were to somehow insert a truly branching list by some error, having the
> properties being functional might well produce very peculiar inferences.  For
> example, if a reasoner were to conclude that _:x and _:y were the same, where
> these were the bnodes of two lists, then the first items in those lists would
> also be inferred to be equal. If the lists were of unequal length then it
> would follow , for example, that first_item of empty_sequence would be
> required to be equal to some resource. I think it would be safer, and better
> engineering, to not assert any properties of those functions, but to rely on
> external checkers to ensure that argument sequences are linear, if that were
> ever an issue that needed checking.
>  SHORT ANSWER:
> This just seems to be complaining that OWL means what it means.  Its semantics
> are open world.  It doesn't have a unique name assumption. If you want to
> represent a closed world notion such as a linear list, you have to close it
> explicitly.  Otherwise, the interpretation is "at least one successor" instead
> of "exactly one successor".  IE, other you are representing a multiway tree.
> Mislabelling a multiway tree as a linear list is going to be unhelpful and
> lead to  confusion in the long run.  Does not depending on checks outside the
> standard for semantics that are in the standard defeats the purpose of having
> a standard?
>
> LONGER ADDENDUM:
>
> I am strongly in favour of external checkers to help users avoid misakes with
> open world reasoners.  We have the same problem with domain and range
> constraints, and the Protege-OWL  tools are providing checkers to make sure
> that users know of potential violations, assent to them, and are able to find
> them as the cause of unexpected classification behaviour, either
> unsatisfiability or reclassification.
>
> More importantly, we provide things to help ensure that the allDifferent and
> disjointFrom axioms are provided where intended. Your examples above only make
> their anomalous conclusions in the face of incomplete axiomization.  Given the
> appropriate allDifferent axioms, they would just report unstatisfiability of
> various concepts. It was not my decision not to use a unique name assumption
> in OWL, but given that OWL explicitly does not have a unique name assumption,
> this is one of the consequences.
>
> If we want to support routine use of such list structures, we will need to
> provide syntactic sugar.   I don't see most users writing things in their full
> form any more than we normally write lists in terms of raw cons cells.  Checks
> will be part of that sugar - just as the sugar for value partitions includes,
> by default, making the property functional and the values disjoint and
> covering. (When users understand enough to over ride the defaults, we have
> achieved something.  In the meantime, these are generally safe.)
>
> Since OWL can clearly express and distinguish notions such as list, binary
> tree, multiway tree etc it seems perverse to use a representation for a
> multiway tree - i.e. "rest_olf_list" without cardinality - to represent the
> much more specific notion of a linear list.  That RDF cannot distinguish them
> because of weaker semantics is hardly a reason not to distinguish them in a
> language with stronger semantics.
>
> In short, I see no reason not to express things in OWL as fully as the
> semantics allows.   Where there are things that cannot be expressed with that
> semantics, then we have need for workarounds and use of syntactic checkers,
> but those will always be outside the standard and subject to error.  Likewise,
> I am all in favour of tools to help people say what they intended and avoid
> errors, but the outcome should conform to the standard semantics where
> possible.
>
>
>> I think the issue of lists of this type in OWL and the inferences that can
>> be made about them probably requires a separate note.
>
>  Since OWL uses the RDF vocabulary and the RDF documentation discusses this
> fairly thoroughly, I wonder if a note is really needed (?)
>
> OWL supports considerable reasoning about lists that RDF does not - as your
> example shows.  It is possible to do some very useful things, but that's not
> the point here.
>
>
>
>> However, I suggest we stick to this much here and do that separately later.
>
>
> Agreed. Pat
>
> Regards
>
> Alan
>
>

--
Alan L Rector
Professor of Medical Informatics
Department of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL: +44-161-275-6188/6149/7183
FAX: +44-161-275-6236/6204
Room: 2.88a, Kilburn Building
email: rector@cs.man.ac.uk
web: www.cs.man.ac.uk/mig
        www.opengalen.org
        www.clinical-escience.org
        www.co-ode.org