dr-shorthair
commented
7 years ago I believe the nomenclature drift issue is a much bigger issue in the life-sciences than the earth sciences, particularly in taxonomy. Of course it would be absurd to claim that there is no semantic drift, but the current scope of SWEET is a higher level than individual 'species', genes, organisms, etc. This might change in future, and there may be branches of SWEET where we should flip to numerically based URIs.
So I would agree with the proposition that 'earth science terminology is more static than in the life sciences and medicine', at least in the current scope of SWEET, so I don't think we need a general switch.
A major issue with SWEET as-is is that there are very few definitions or even labels on classes, so the last fragment of the URI does double duty, and that along with the position in the hierarchy/graph is mostly all we have for 'definition'. I raised this in another ticket #20
graybeal
cmungall
pbuttigieg
lewismc
brandonnodnarb
rduerr
smrgeoinfo
rrovetto
Continued from #48 - if a change is URIs is under consideration then I want to make the case for OBO-style URIs, or some other meaning-free-identifier scheme, e.g. UUIDs or URIs with incrementally assigned numeric fragments.
The reason we do this in almost all biological ontologies (whether OBO or not) is that the URI is associated with the meaning/concept, which remains static, while nomenclature drifts. For example:
This has clear practical advantages for biocurators and databases using the ontology. For example, a genome database may want to describe genes in terms of their biological functions using the Gene Ontology. This will be stored as associations between gene IDs and GO IDs. For both IDs, the curator makes a decision based not on terminology, but based on the defining features of the gene and the function. For the gene this may be position on the genome, sequence etc. For the function, this is based on the definition provided in GO. Both the gene ID and the GO ID are numeric, and never change (although IDs can be obsoleted, with an audit trail retained). However, the terminology and nomenclature is frequently in flux. If the database stored the association between IDs that had the nomenclature embedded, then either the curator would need to constantly update the records, or the nomenclature would need to be frozen, which would create all kinds of problems.
It's not totally clear what the intent is with SWEET. If the goal is to provide an axiomatized dictionary, then the current strategy of meaningful URI fragments is appropriate. The class really does represent the string. This is also consistent with the way SWEET creates equivalence axioms between named classes (in contrast to OBO where each concept is represented by a single class and alternate terminology is represented as annotation assertions).
Also, it may be that earth science terminology is more static than in the life sciences and medicine, in which case this ticket can be easily ignored.
However, if the classes represent different concepts or language-independent things, you may want to consider decoupling the concepts from the terminology.
Of course, meaningless URIs are not without their downsides. It's much harder to author or read triples without an IDE or appropriate interface. In practice this is less of a hurdle for domain experts, who will not be working with raw turtle in any case.