I'm drawing together Kevin's post with one posted by Hilmar Lapp some time ago on the same subject. Both are included below.

Despite the fact that Dublin Core does actually use the hasDomain property for some terms (admittedly very few), I believe that the hasDomain property has been incorrectly applied in DwC as it currently stands. The definition of hasDomain from Dublin Core is "A Class of which a resource described by the term is an Instance." As Hilmar points out (below), using the domains as currently defined in DwC terms, a resource described by a dwc:collectionCode would be an instance of a dwc:Occurrence (the hasDomain property of dwc:collectionCode is http://rs.tdwg.org/dwc/terms/Occurrence). Similary, a resource described by a dwc:scientificName would be an instance of a dwc:Taxon. Clearly we will want to describe Occurrences with dwc:scientificName without asserting that the Occurrence is a dwc:Taxon, so the domain assignment is too constraining.

The intention in DwC was much more loose than this - the domains were set simply to logically group the properties, not to assert instances of classes. Yet, the groupings in the Darwin Core documentation and schemas are extremely useful. I propose that we maintain those informal associations, but remove the hasDomain assignments so that misleading ontological assertions are not made.

Now, getting to Kevin's agent conundrum. If we use a common agent property and have no domains for the terms how would we distinguish specific roles? Instead, the specific roles are elaborated in their own terms that all refine http://rs.tdwg.org/dwc/terms/accordingTo (see http://rs.tdwg.org/dwc/terms/history/index.htm#accordingTo-2009-01-21), the definition of which is "Abstract term to attribute information to a source." This isn't the same as dcterms:source or dcterms:Agent, though if the distinction seems too subtle, each of the roles could be amended to be given http://purl.org/dc/terms/Agent as the value of the hasRange property.

About the ontology, again, until there is active work in that area, I think the pragmatic way forward is to let it be the burden of the ontology to make use of terms from actual standards.

On Sun, Jul 12, 2009 at 5:05 PM, Kevin RichardsRichardsK@landcareresearch.co.nz wrote:

I'm not sure if this is my misunderstanding of RDF and ontologies, so correct me if I am wrong...

I noticed that the http://rs.tdwg.org/dwc/terms/geoReferencedBy property has domain http://purl.org/dc/terms/Location. Is this the standard way of working with Dublin Core? I would have though it would be better to define a "common" person/agent type property (or better still reuse FOAF or DC Agent types - http://purl.org/dc/terms/Agent).  This property has no range and can therefore be applied to any RDF class/element.  This would remove the need for repetitive elements that refer to Agent type properties, eg geoReferenceBy, identifiedBy, measurementDeterminedBy, etc.

This was part of the intention of the TDWG Common Vocab. (http://rs.tdwg.org/ontology/voc/Common.rdf), which brings me to my next question - how do the DwC terms relate to the TDWG ontology? - do we need a "mapping" to them?  I suppose, because the DwC ontology needs to be ratified as a standard, it can't really rely on an un-ratified ontology - but it would be good to specify how the two should work together, or map to each other.

Otherwise I think this is looking excellent - a fantastic step towards an overarching ontology for TDWG data objects.

Kevin

----- On Jun 5, 2009, at 9:20 PM, John R. WIECZOREK wrote:

[...] The first s organization of the standard. The domains help organize the properties in ways that help people to understand their meaning and purpose.

Hilmar replied: That's a worthwhile goal but isn't that using the wrong means? I.e., specifying the domain is telling something - and in an unambiguous manner - to machines, not humans.

You can of course create user interfaces or renderings that turns the domain specification into something meaningful to humans, too. But you could use any term property for that, couldn't you? I.e., you could use a property that doesn't imply specific and possibly far reaching machine inferences. (I haven't had time to check yet but I suspect that there are - W3C or not - conventions for how to add human-targeted class and property annotations to vocabularies.)

(BTW for comparison purposes, note that DC doesn't assert any domain or range for any of its terms, allowing the broadest possible use.)

On Jun 5, 2009, at 9:20 PM, John R. WIECZOREK wrote: [...] The second reason for the domain assignments is that we lack a formal ontology, and this is an attempt to have one to govern at least the terms within this standard.

Hilmar replied: I'd argue that it's worth distinguishing between a metadata vocabulary and a formal ontology, and that having one product try to be both may limit its ability to fully satisfy both.

A standard metadata vocabulary with the primary purpose that we all call the same thing by the same name is broadly useful and can help enormously with data integration across fields as diverse as genetics, genomics, systematics, ecology, and taxonomy (and more, as we heard in London). A formal ontology that supports inferences over integrated data is also highly useful, but not necessarily at the same breadth.

On Jun 5, 2009, at 9:20 PM, John R. WIECZOREK wrote: What is the potentially problematic future case of asserting that a specimen is an dwcterms:Occurrence? It is one.

Hilmar replaied: My example was indeed a weak one, as we are indeed referencing specimens. For example, I couldn't use dwcterms:collectionCode to assert a code for a museum collection, because it would imply that the collection is an dwcterms:Occurrence.

Maybe you don't want people to use DwCTerms for anything else other than describing specimen records, but I'd argue that without such limitations the standard could become much more broadly useful.

As a comparison, dc:creator and dc:title are meanwhile being used in vastly more contexts than the original authors of DC had probably imagined. I'm not sure this would have also happened if applying dc:title implied specific assertions about the nature of what it is being applied to.

Just my $0.02.

-hilmar