Hash: SHA256

Am 31.05.2016 um 21:02 schrieb Adrian Pohl:

> <Instance> lv:inSeries [
>         a lv:SeriesRelation ;
>         lv:series <> ;
>         lv:numbering "v. 30, pt. 3 = Bd. 30, Teil 3" ] ;
>     lv:inSeries [
>         a lv:SeriesRelation ;
>         lv:series <> ;
>         lv:numbering "v. 2357" ] .
> Note that the bnode entity rather is a _relation_
> (lv:SeriesRelation) of an instance to a series and is not itself a
> series.

There seems to be a strong tendency of avoiding blank nodes for
almost whatever price, probably from a data management perspective:
[If I want to modify that statement (or inject it into a different
database) in isolation it has to be identified. But elevating the
complex statement to an identifie[re]d entity raises the question
of suitable constraints and class membership]

RDF (independent of any serialization syntax) is not very expressive
when it comes to relations, and not everything is a case of RDFS
subproperties or meta-data injected into the object of a relation
(like who crafted this identifier or who is in charge of the target
I'm increasingly under the impression that /most/ of our relations
tend to be qualified, e.g. we not only want to state this is
"partOf" that but furthermore provide an answer to the question
"/what/ part of that?". (Similarily when we state an "author"-
relation to an entity we could like to record how this normalized
relation is expressed by the resource itself - not only in the
rather coarse way of the complete statement of responsibility. Or
look at MARC where almost every non-trivial fact (including
relations) might be qualified by $2 and/or $5's: For whom and
under what constraints was this relation recorded.

We have encountered many situations on this list where some
theoretically "direct" relation between entities had to be
kind of encapsulated into some subgraph and if I recall correctly
there always has been some queasyness as to where this
additional level of indirection should belong to (like an
"aspect" of the source graph or some extension of the
target graph). As I understand Adrian's point of view
it might look like a subgraph of the source graph (anything RDF
always looks like that...) but this subgraph does not represent
aspects or parts of neither source nor target but it rather
*is* (or records) the relation between the two and therefore
really has a very different nature from the "usual" entities.

So perhaps we should simply abandon the conception that any
relation between S and O must be expressible by a /simple/
statement <S> <rel> <O>.

viele Gruesse
Thomas Berger
Version: GnuPG v2
Comment: Using GnuPG with Thunderbird -