More on relationship of ontologies to information modelling.

[Thomas Beale]

I posted here a couple of months ago on this topic; Werner responded with various interesting references and discussion points.

I was somewhat derailed from further discussion by job and house move events, so did not get to resolve things.

I thought I would post here a diagram and an explanation of the general problem, which is: what is the relationship between realist ontologies like BFO, and information models and systems that digitally represent (snapshots) of real world individual entities described by such ontologies.

The problem is the relationship among four entities: a) realist ontologies (as governed computable artifacts); b) real entities described by the ontologies (actual machines etc) and c) information models (also governed entities) and d) data instances of c).

The challenge is that we (unavoidably) have information instances representing both individuals, e.g. a single ultrasound machine with a UDI, and also machine 'models', i.e. types of machines (or other manufacturable artifacts). This enables the data instance for 20 machines of the same type in use in a particular provider to point to a record representing that machine model, which provides the common model-level data, e.g resolution, image sample rate, or whatever. The individual machine records have more data points of their own, e.g. maintenance history, cleaning, software / hardware additions and so on.

It's not even just instance v model. There are model families, and common core technologies, so there is generally a hierarchy of progressively more general artifact types sitting above any actual machine in the real world. The diagram shows 2 levels of this. The hierarchy of 'models' poses some challenges for ontologies, possibly manageable in the same way as for biological categories above species. Example: there might be 3 levels of technology just within the 'Sonograph' level' at Philips, but only one at Agfa.

In the diagram, I have shown two classes in the information model - Artifact_kind, and Artifact (corresponding to a category Artifact I added below BFO). Since we want digital records both of instances (Ultrasound machine with UDI 1234, up on floor 5), and potentially a full hierarchy of machine models / families etc, that enable common characteristics to be 'inherited' (informationally), we need both of these classes. This is not immediately apparent from an ontology like BFO, which provides categories like Object, but which are usually understood to stand for kinds. BFO doesn't distinguish instantiable kinds from 'abstract' (non-directly-instantiable) kinds.

There is a confusing English term that is problematic here: 'instance of'. In common parlance, we use that to mean 'Davey Jones' instance-of human (ontic entity -> ontology entity), and also data record 1425 (representing Davey Jones in the hospital admin system) instance-of class Person in some DB schema or UML model (data instance -> information model class). In the diagram, I have use 'instance-of' to mean what it means in BFO, and 'data-instance-of' to indicate data being an instance of its formal type.

So the question is how ontologies describing real world things relate to information models that define information records of those real things.

Currently I think the lack of methodology to answer this question is one of the factors preventing more widespread use of BFO and other related ontologies in information systems development. I also think it has led to IAO, which I think is not practically useable.