BFO DL-Query Oddities noticed by Amanda Hicks
Opened this issue · 3 comments
From alanruttenberg@gmail.com on March 01, 2013 16:17:10
Amanda Hicks, UAMS, aellenhicks@gmail.com
I have noticed that there are some unsound results for DL queries in BFO. In short, some of the queries of the form R only x produce strange subclasses. I first noticed this when a query on one of our ontologies (Apollo-sv) produced Acquired Immunodeficiency Syndrome as a descendent of simulates only continuant AND simulates only occurrent. I think I have traced this oddity back to BFO.
Below is a set of notes I took on the queries that I made in case they're helpful. I'm not sure where the problem lies though, but if you have any questions about these queries, let me know. And if you make any headway figuring out where the problem is, I would be really interested in hearing about it!
In each of the ontologies below I created a dummy relation R and restricted the domains or ranges in various ways. There are some unsound subclass relations that result in some queires of the form R only x. They do not result when some is used.
BFO 1.1 http://bfo.googlecode.com/svn/releases/1.1/bfo.owl and
BFO 1.1.1 http://bfo.googlecode.com/svn/releases/1.1.1/bfo.owl Relations with a domain on the branch of continuants generate odd DL query results when the query is of the form R only x.
The general pattern appears to be that R only x subsumes the compliment class of the domain of R for all x. However, I have not tried all combinations.
Here are some sample queries:
Domain: continuant
Query: R only continuant
Subclass: occurrent
Query: R only dependent_continuant
Subclass: occurrent
Query: R only occurrent
Subclass: occurrent
Based on similar queries it is probably the case that when the domain of R is a continuant, R only x always yields occurrent as a subclass.
Domain: dependent_continuant
Query: R only continuant
Subclass: independent_continuant, occurrent, spatial_region
Based on similar queries it is probably the case that when the domain of R is a dependent_continuant, R only x always yields these subclasses.
Domain: generically_dependent_continuant
Query: R only continuant
Subclass: independent_continuant, occurrent, spatial_region, specifically_dependent_continuant
Based on similar queries it is probably the case that when the domain of R is a generically_dependent_continuant, R only x always yields these subclasses.
Domain: specifically_dependent_continuant
Query: R only continuant
Subclass: independent_continuant, occurrent, spatial_region, generically_dependent_continuant
Based on similar queries it is probably the case that when the domain of R is a specificially_dependent_continuant, R only x always yields these subclasses.
Domain: independent_continuant
Query: R only continuant
Subclass: dependent_continuant, occurrent, spatial_region
Based on similar queries it is probably the case that when the domain of R is a specificially_dependent_continuant, R only x always yields these subclasses.
Ruttenberg BFO 2 http://bfo.googlecode.com/svn/releases/owl-ruttenberg-2010-05-25/ruttenberg-bfo2.owl Here are some sample queries:
Domain: continuant
Query: R only continuant
Subclass: none
Query: R only dependent_continuant
Subclass: none
Query: R only occurrent
Subclass: none
Based on similar queries it is probably the case that when the domain of R is a continuant, there is no problem.
Domain: dependent_continuant
Query: R only continuant
Subclass: none
Based on similar queries it is probably the case that when the domain of R is a dependent_continuant, there is no problem.
Domain: generically_dependent_continuant
Query: R only continuant
Subclass: none
Domain: generically_dependent_continuant
Range: independent continuant
Query: R only continuant
Subclass: entity
Domain: n/a
Range: independent continuant
Query: R only continuant
Subclass: entity
Domain: n/a
Range: occurrent
Query: R only continuant
Subclass: none
Domain: n/a
Range: dependent continuant
Query: R only continuant
Subclass: none
Domain: n/a
Range: material entity
Query: R only continuant
Subclass: none
Query: ‘g depends on’ only continuant
Subclass: entity
Query: ‘is concretization of’ only continuant
Subclass: entity
Query: concretizes only continuant
Subclass: entity
Based on similar queries it is probably the case that something funny is happening on the continuant branch.
BFO 2
2012-11-15-bugfix http://bfo.googlecode.com/svn/releases/2012-11-15-bugfix/owl-group/bfo.owl There are a multitude of classes that are subclasses nothing. In addition, it seems that BFO 2 has the same problem as BFO 1.1 and BFO 1.1.1
The general pattern appears to be that R only x subsumes the compliment class of the domain of R for all x. However, I have not tried all combinations.
Here are some sample queries:
Domain: continuant
Query: R only continuant
Subclass: occurrent
Query: R only generically-dependent_continuant
Subclass: occurrent
Query: R only occurrent
Subclass: occurrent
Based on similar queries it is probably the case that when the domain of R is a continuant, R only x always yields occurrent as a subclass.
Domain: generically_dependent_continuant
Query: R only continuant
Subclass: independent_continuant, occurrent, specifically_dependent_continuant
Based on similar queries it is probably the case that when the domain of R is a generically_dependent_continuant, R only x always yields these subclasses.
Domain: specifically_dependent_continuant
Query: R only continuant
Subclass: independent_continuant, occurrent, generically_dependent_continuant
Based on similar queries it is probably the case that when the domain of R is a specificially_dependent_continuant, R only x always yields these subclasses.
Domain: independent_continuant
Query: R only continuant
Subclass: dependent_continuant, occurrent, generically dependent continuant
Based on similar queries it is probably the case that when the domain of R is a specificially_dependent_continuant, R only x always yields these subclasses.
Original issue: http://code.google.com/p/bfo/issues/detail?id=153
From alanruttenberg@gmail.com on March 01, 2013 13:18:07
We shouldn't bother with the ruttenberg-bfo2 ones, since that isn't in the line of development. So priority: BFO2, BFO1 as time permits
From steschu@gmail.com on March 01, 2013 13:45:18
The expression
simulates only continuant AND simulates only occurrent (1)
prevents that anything is in the range of simulates, because continunant and occurrent are disjoint
This expression is equal to
simulates only (continuant and occurrent) (2)
and because the intersection (continuant and occurrent) is equal to nothing
simulates only (Nothing) (3)
which can be syntactically be transformed into
not (simulates some Thing) (4)
Does this explain the inference?
To your more basic example:
Domain: continuant (5)
Query: R only continuant (6)
Subclass: occurrent (7)
This can be explained as following:
R: domain continuant (5)
is in DL by definition equivalent to the axiom
R some Thing subclassOf Continuant (8)
R only Continuant (6)
can be transformed into
Not (R some (Not Continuant)) (9)
which implies
Not (R some Thing) (10)
because every satisfiable expression implies Thing
As R some Thing is a subclass of continuant (8) then (10) cannot be a subclass of continuant
It is eqivalent to
Entity and (not continuant) (11)
which subsumes occurrent (7)
q.e.d.
From cmung...@gmail.com on March 09, 2013 14:22:27
I suggest closing this item. There is no bug here in any version of BFO. This is how we expect "only" restrictions in OWL to work.