Susie asked me to put together a list of potentially contentious issues. Here are few that I thought of.
- Purpose. The recommendations document is in effect a plea for others to publish RDF according to certain "recommendations", and an agreement to work with what each of us publishes assuming we follow those same recommendations. Do others agree? - Univocality. I would like to recommend that a URI is not "good enough" for our purposes if it is used to denote two things. I know that this technique (e.g. using the same URI for a resource and its metadata) has been defended in the past (maybe not inside HCLS). - Well-definedness. I would like to recommend that a URI is not "good enough" if it is not well defined. Some say that URI's without a clearly defined "meaning" (referent and/or rules of use in RDF) can't be used in semantic web applications. Other say, or imply, that they can. - Avoid speculation. I would like to recommend that uses of URI's be based on rules, not heuristics. However, some people recommend e.g. the "follow your nose" approach to metadata harvesting, which is heuristic. Content negotiation also falls in this category. - Opaqueness of URIs. Does it matter at all? If so, what's important - URN vs. HTTP, etc.? If not, how does resolution work? W3C recommends use of HTTP URIs and wants URIs to resolve (to 200's for IRs, 303's for non-IRs). OTOH some people seem to still like URNs, either in spite of the fact that they don't resolve, or maybe because they don't. Should "good enough" URI's reveal anything about their denotations? - URIs for public resources. In order for us to work together we clearly need to share URIs for things that we talk about, e.g. Entrez database records. The devil is in the details: how do the get URIs defined and documented, how do we ensure that there's no punning (e.g. gene record vs. gene), who is the authority for the URIs, how do they resolve (if at all). Only after we answer all of these questions should we decide what the exact URI's should be. - Shared resolution rules. Do we need this level of abstraction? (The answer may depend on answers to other questions.) If so, how should the rules be written - in RDF maybe? - Versioning. What is it for, what does it mean, how is it expressed?
