AHG3 recommendation re ISO 19109 - various

[PeterParslow]

This standard is also specific to UML, but also includes the General Feature Model (GFM) as one of the metamodels

• separate GFM from the "Rules for Application Schema"
• show how the GFM can be applicable in OWL - this may involve making some changes to the GFM, or at least to the detailed UML expression of it. Essentially, we want the "top level" GFM to be equally implementable in OWL & UML; if a UML flavour/profile can support lossless creation of OWL, then that could be a way.
• clarify (here or in ISO 19103) the distinction between "concept modelling" (or "conceptual modelling") and logical data modelling. Note: the OGC Architecture Board is actively looking at this.
• Discuss the meaning of Application Schema (focus on logical model, data specifications https://isotc211.geolexica.org/concepts/15/) versus Conceptual model (focus on 'pure' semantics, information specifications). Note that TC 211 use of "application schema" is (currently) specific to "UML model of an application"
• discuss the place of "instance models"
• discuss the question of instances that belong to more than one feature type. (Note: this relates to the current question issue in ISO 19103/19109 vs UML, as to whether specialised classes default to disjoint or not.)
• Possibly discuss handling n-ary relationships & the fundamental structure of relations (same as, same place as, 'converse')

[PeterParslow]

Also asked in AHG3 report:

Is it actually helpful to draw a clear distinction between the "instance" level and the higher levels? Graph data does not require this distinction: for example a "type definition" is just another data instance. However, in practice graph implementations - such as OWL and RDF do distinguish "class"; philosophers have long considered (e.g. from Aristotle on) that this represents a fundamental way that human brains work, and it also gives efficiencies in computer processing. That said, we need to recognise that there are cases where any classification will fail, and 'open world models' handle those cases better.