[Crm-sig] Recording absence
thanasis at softicon.co.uk
Tue Nov 5 16:50:59 EET 2019
Following the Linked Conservation Data workshop and the last SIG in
Crete I am summarising the problem of documenting non-existence.
An example of non-existence is: a book cover (a particular) without
tooled decoration (a type).
Options for encoding:
1) As discussed here:
we could have a new E55 Type "books without decoration". This is a good
solution but the problem is that we will need an unmanageable number of
composite thesaurus terms to cover all possibilities, e.g. things
without a feature, or types of events which did not happen etc.
2) In past SIGs we have mentioned negative properties. This is also a
good solution but not quite in scope. A negative property requires
particulars for domain and range. So I can say that:
cover(E22 Man-Made Object) → NOT P56 bears feature → tooled
decoration(E25 Man-Made Feature)
This would mean that the specific book does not carry the specific
decoration. But I want to say that the specific book does not carry any
3) So I pestered Carlo for a few days and he says:
"To express negative information in an ontology, it is recommended to
use specific axioms. For example, to state that certain books have no
decorations the axiom would require to create a special class for those
books and to make that class a sub-class of the class expression
'individuals with less than 1 decorations'. This will require a class
and an axiom to be created for each type of negative information to be
expressed. But it has the advantage of using a standard OWL 2 DL
inference engine to reason about that negative knowledge, both for
maintaining consistency and for query answering."
So what Carlo thinks is that option 1 is reasonable and in fact instead
of using simply a thesaurus, one should elevate these definitions to
ontology classes and axioms.
I would be interested to hear views from the list, as I am not sure how
to model such statements. Those of you who have looked at this in the
past, do you get a sense of the scale for negation statements?
P.S. A parallel thought which did not capture Carlo's imagination was a
"typed negative property", i.e. create new negative properties with E55
as range as in:
cover(E22 Man-Made Object) → NOT P56 bears feature of type → tooled
but I am not sure how this would translate to logic in an inference engine.
More information about the Crm-sig