Another modeling question
Larry Hunter
One more thing: I would like to say both that a footnote is part of a
document, and it is about different part of that same document. How
do I do that?
Larry
Alan Ruttenberg
are about themselves.
See http://www.w3.org/TR/2009/CR-owl2-syntax-20090611/#Self-Restriction
Attached is a small ontology that demonstrates this. However I don't
get consistent behavior among the three reasoners when using this in
pellet4 so I will file a bug report and add it to the OWL 2 tests.
-Alan
>
> Larry
>
> >
>
Bjoern Peters
footnote part_of some document
and
footnote is_about some information content entity part_of some document
and wants to assert that both 'document' statements refer the same one. I posed the same modeling question to you (=Alan) before, and believe that you said this was not possible.
- Bjoern
Alan Ruttenberg
> I believe Larry is trying to say that:
>
> footnote part_of some document
> and
> footnote is_about some information content entity part_of some document
>
> and wants to assert that both 'document' statements refer the same one. I posed the same modeling question to you (=Alan) before, and believe that you said this was not possible.
being about a *different* part.
However, I'm not sure it would be correct to say that footnotes are
always about the document they are in. Certainly the footnote marker
is - it refers to the place in the document where the footnote is
made. However footnote are often about something the article is about,
but necessarily the article.
Regarding making the statement, anyways, I'll have to think about it.
As you point out, its the sort of thing OWL isn't typically good at,
but maybe there's a trick with property chains.
-Alan
Larry Hunter
and wants to assert that both 'document' statements refer the same one. I posed the same modeling question to you (=Alan) before, and believe that you said this was not possible.
As you point out, its the sort of thing OWL isn't typically good at,
but maybe there's a trick with property chains.
Alan Ruttenberg
OWL is a restricted portion of FOL, and is mostly 1 variable
quantification. The two variable (and more) quantification only comes
in very restricted forms - axioms about properties such as
SubPropertyOf, some properties of properties, such as expressing that
a property is transitive, symmetric, functional, etc, and and in
property chains. Use of property chains have serious constraints.
http://www.w3.org/TR/2009/CR-owl2-syntax-20090611/#Global_Restrictions_on_Axioms_in_OWL_2_DL
The restrictions are taken in the interest of keeping OWL decidable. FOL isn't.
As an example, it isn't, for such reasons, possible to express, in OWL
2, the class of all cyclic structures - namely all those structures
that are self-related by a transitive property such as connected-to,
even though you can express this in FOL.
-Alan
> Larry
Larry Hunter
On Jul 29, 2009, at 9:27 PM, Alan Ruttenberg wrote:
> The restrictions are taken in the interest of keeping OWL decidable.
> FOL isn't.
Decidable for subsumption inference -- not the most interesting
question in the world. *grumble*
Is this really that hard, even in OWL 2?
Larry Hunter
On Jul 29, 2009, at 9:27 PM, Alan Ruttenberg wrote:
>> Larry Hunter wrote:
>> It must be possible. It would be trivial FOL: footnote(x) =
>> part_of (x, y)
>> ^ document (y) ^ is-about (x, z) ^ part_of (z, y) ^ != (x, z)
>
> As an example, it isn't, for such reasons, possible to express, in OWL
> 2, the class of all cyclic structures - namely all those structures
> that are self-related by a transitive property such as connected-to,
> even though you can express this in FOL.
But there's no self-reference in my FOL definition of footnote -- just
a 'not equals' assertion.
DL is, as ever, an incomprehensible to me as choice for a restriction
on a representation language. If it can't even represent footnote,
who cares that subsumption is decidable? Isn't OWL 2 more like OWL
full in this regard (I hope)?
Larry
Alan Ruttenberg
>
> On Jul 29, 2009, at 9:27 PM, Alan Ruttenberg wrote:
>
>>> Larry Hunter wrote:
>>> It must be possible. It would be trivial FOL: footnote(x) = part_of (x,
>>> y)
>>> ^ document (y) ^ is-about (x, z) ^ part_of (z, y) ^ != (x, z)
>>
>> As an example, it isn't, for such reasons, possible to express, in OWL
>> 2, the class of all cyclic structures - namely all those structures
>> that are self-related by a transitive property such as connected-to,
>> even though you can express this in FOL.
>
> But there's no self-reference in my FOL definition of footnote -- just a
> 'not equals' assertion.
What's on the side of the != assertion :)
>
> DL is, as ever, an incomprehensible to me as choice for a restriction on a
> representation language. If it can't even represent footnote, who cares
> that subsumption is decidable? Isn't OWL 2 more like OWL full in this regard
> (I hope)?
Representing is easy. Making sure you can inference correctly based on
what is said is another thing entirely. Its true that OWL 2 make a
certain set of choices. Mostly they have worked (well enough) for us
so far.
Going forward I expect, following Chris' lead, we will be at least
capturing those axioms that can't be expressed in OWL as FOL using
some syntax of common logic, so we can express what we mean, even if
we can't compute with it adequately yet.
-Alan
>
> Larry
>
Chris Mungall
On Jul 29, 10:22 pm, Alan Ruttenberg <alanruttenb...@gmail.com> wrote:
adequately with it. For example, Larry's fragment of FOL is computable
using relational databases (assuming some trivial pre-computations
such as calculating the closure of part_of) which should scale to
billions of instances. I'm guessing Larry has a large corpus. For a
smaller corpus there are other non-OWL fragments of FOL for which
efficient decision procedures exist.
There is no one-size-fits-all solutuion. DLs excel at calculating
subsumption between classes (module cycles), which means they are
primarily used during the ontology development cycle but rarely
outside it (at least AFAIK in bioinformatics). I'm hoping we see more
applications of scalable reasoning with ABoxes in the near future, but
until we do it's neither the case that OWL is computable for all
problems nor that non-OWL is non-computable.
Cheers
Chris
> -Alan
>
>
>
> > Larry
Alan Ruttenberg
I'm hoping we see more
> applications of scalable reasoning with ABoxes in the near future, but
> until we do it's neither the case that OWL is computable for all
> problems nor that non-OWL is non-computable.
however, that we are going to have to be clever and strategic about
how we manage inference.
In any case, it is worth pointing out that an sufficiently small
fragment of FOL is computable. The tricky stuff happens when
expressivities are combined. DL represents one design point - maximum
expressivity for sound and complete reasoning that is decidable. There
are profiles of OWL that offer less expressivity, and better
guarantees on performance, or which compromise completeness, and there
are extensions that add elements of FOL not expressible in OWL that
allow for less complete or undecidable reasoning.
-Alan
>
> Cheers
> Chris
>
>> -Alan
>>
>>
>>
>> > Larry
> >
>