[Attempto] News from EulerGUI; Ace and Ace View ontologies repository

[Kaarel Kaljurand]

Hi,

On Tue, Apr 6, 2010 at 10:33, Jean-Marc Vanel <jeanmarc.vanel at gmail.com> wrote:
>>
> It's clear that OWL is expressive enough to capture the large majority of
> controlled language snippets.

You are right, the ACE View and AceWiki examples do not contain any
SWRL-only sentences. The reason is not so much that OWL is expressive enough,
simply we have tested these tools mainly with OWL as there haven't
been any SWRL reasoners around.

> But how would you characterize the snippets that are translated to SWRL ? In
> general, in the publications, there is little details on SWRL.

If DRS->OWL fails then DRS->SWRL is tried. The sentences where the first fails
and second succeeds feature complex anaphoric references which cannot be
expressed using relative clauses. E.g.

(1) If a dog chases a cat then the cat runs-up a tree.

can be expressed as

(1') Every cat that is chased by a dog runs-up a tree.

where no "the NPs" remain, so this is OWL-compatible, in
Manchester Syntax

(1'') cat and some inv(chase) dog SubClassOf some run-up tree

However, a sentence like

(2) if a dog catches a cat then the dog bites THE cat.

cannot be rewritten so that the need for both the explicit
references "the dog" and "the cat" disappear.
So this is converted to SWRL (which allows explicit variables):

(2'') dog(X) AND cat(Y) AND catch(X, Y) -> bite(X, Y)

> As side note, for the Geo ontology, the snippet
> If X borders Y then Y borders X.
> is translated as
> border subPropertyOf inv(border)
> I would have expected border to be a symmetric property.

It is a symmetric property, just not expressed using the dedicated
syntactic construct that OWL provides for symmetry. Note that:

P SubPropertyOf inv(P)

is semantically equivalent to

SymmetricProperty(P)

For more of such equivalences, see Table 1 in the publication:

Verbalizing OWL in Attempto Controlled English
Kaarel Kaljurand and Norbert E. Fuchs

Since the Attempto DRS language does not provide a dedicated syntax
for expressing symmetry, it is more straight-forward not to generate
the OWL's symmetry axiom in this case. I agree, however, that targeting
a richer syntactic subset of OWL would have some value as this would
improve compatibility with OWL tools that closely reflect the syntactic
sugar of OWL (e.g. Protege has a "checkbox" for every OWL property
axiom type).

> How do you test the regressions for the ACE ==> OWL+SWRL ?
> There must be some SWRL cases there.

Yes, I forgot before, there are some 200 ACE->OWL/SWRL testcases.
The regression testing
simply works by generating the OWL/SWRL forms for each of the test cases,
saving them into a file, and running a Unix diff against an older run. (Not so
clean to make it officially public but works. ;))

I've included a Prolog file with the test script and the tests in the
attachment.
It's not that easy to extract the SWRL tests from there but maybe it helps.

--
kaarel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_owlswrl.pl
Type: application/x-perl
Size: 16416 bytes
Desc: not available
URL: <http://lists.ifi.uzh.ch/pipermail/attempto/attachments/20100407/7adfb7a0/attachment.bin>