On Thu, Jan 31, 2013 at 4:18 PM, Magnus Enger <[log in to unmask]> wrote:
The question does raise another interesting problem though: As far as I can tell, moving from MARC to RDF should mean that the concept of the "record" as an island unto itself is dead - the future is built up of descriptions of "things" that are linked to each other, in interesting ways. One challenge might be: If we want to "copy" a description of a "thing", how far should we follow the links it is involved in? When have we harvested enough of the context to have a good description of the "thing"?
The semantics of a bibliographic record are quite subtle.
- A record does not contain every known assertion about a described thing.
- There can be many different descriptions of the same thing.
- Records may be derived from other records; however, each record is a single set of statements, with a single claimed provenance which applies to every statement in the record.
- These descriptions may make statements about objective facts that are inconsistent in such a way that at least one statement must be incorrect (for example, pagination).
- These descriptions may make statements that are to some degree subjective and where all such statements may be correct (for example, "the main subject of the work").
- Many statements from different descriptions can be combined to give a unified description.
- Some descriptions may contain sets of statements that are only correct when taken as a whole, and which, if combined with corresponding statements from a different record, entail results that no creator of any of the records intends or would accept. For example, this can often be the case if FAST headings are used.
A record roughly approximates to a graph in SPARQL where each statement has identical provenance information.