Meeting minutes
Issue 76: Ordered lists (Leader: Jim Balhoff)
https://
jim: Can we eliminate option 1, because it does not retain duplicate items in the list?
jim: Actually, from the redundant ordered list, you could reconstruct all of the items.
jim: Odd to have alias have both sometimes a value and sometimes a list.
gopi: Can we use RDF containers for FHIR ordering?
david: We could, but they are very inconvenient to use in SPARQL.
david: The fhir:index allows SPARQL queries to use ORDER BY fhir:index
david: Using OLO might not be much benefit over FHIR R4, because it's pretty similar, but using olo:index instead of fhir:index.
jim: Is the redundancy worthwhile?
Action: Jim to transfer his notes into the issue list.
david: Any other options we should consider?
(none mentioned)
jim: Anyone want to keep option 1?
emily: Unsure of the implications in practice.
… Identifier variables can have multiple entries -- like 12 identifiers.
… If you need to be able to use them, are they more difficult to use if they're in an RDF list?
jim: you can use a property path to get items from the list, but you cannot get the index for it.
gopi: In option 0, if an index is skipped, will it cause a problem?
david: I think the spec might already say that the fhir:index's must be in order starting from 0, but I'm not certain. If no, we should say that.
Issue 69: Shorten dot-compound property names?
https://
gopi: The domain and ranges would change if we shorten to just fhir:severity
eric: In pizza tutorial, they recommend calling the property ":topping" rather than specifically ":pizzaTopping" or ":dessertTopping". They can have different ranges but still the same general concept.
… I think the same idea applies here, for both domain and range.
… Also, domains are typically considered harmful except for a particular task. But if you put it in your general ont, then it prevents people from using it for other purposes. So general advice is to avoid domain and range restrictions.
eric: We already have proof that short names can be unambiguous, because they're already used in JSON and XML.
… The only issue is that it might confuse definitions, e.g., tltle for name of a work, vs an honorific.
… Nothing in the FHIR pub process catches those. They've renamed a bunch of properties to avoid that.
… If we find properties with the same name but radically different definitions, there will be an incentive to resolve that even aside from RDF.
david: I like that argument. Very compelling.
gopi: What about some very long ones? e.g., ObservationDefinition.qualifiedInterval.category
david: That means that the definition will not depend on its context of use. Might that lose a lot of useful information?
… Where would it go?
eric: It would be in the shex though.
… You could also do it in OWL class.
david: Yes, we would want to generate it.
david: Next steps?
eric: Daniel is working on NLP to extract the core definitions for short properties.
… Need to wait for those results before making more progress on this issue.
ADJOURNED