I've responded in-line to a couple of the excellent points Thomas Berger makes.

A. Soroka
The University of Virginia Library

On Jul 27, 2014, at 6:41 AM, Thomas Berger <[log in to unmask]> wrote:
> Am 26.07.2014 02:17, schrieb Tennant,Roy:
>> [snipped]
>> Semantics should not be embedded into text strings.
> True. As long as we talk about machine representations of something. When it comes to humans consuming that data the picture may be different:

This is such a good point. It's important to remember in our conversation about machine-oriented data that we expect to view it and manage it with the aid of software. The fact that some particular use may not be immediately available unaided from it is not, in and of itself, a problem.

> "Thomas Mann's Zauberberg" or "Der Zauberberg / by Th[omas] Mann" usually give sufficient information without bothering us with labels like "title proper" or "statement of responsibility" which I'm forced to understand first and match against my internal concepts secondly, thirdly concluding that they are compatible enough or don't fully understand and don't care anyway.

This sounds like very bad software, which does not do a good job of creating a human-appropriate representation from machine-orineted data. Using screen labels like "title proper" without at least supplying context and assistance does sound confusing. But it doesn't seem to me to say anything about how well or poorly the machine-oriented data is designed.

> Also <gndo:forename>Zedong</gndo:forename><gndo:surname>MaĆ­</gndo:surname> does not imply any order or prescribes a delimiter, therefore a variantNameForThePerson cannot be constructed from a variantNameEntityForThePerson (at least without additional rules to be known by applications).

I think this is often a good thing. It seems to me that the order (and possibly the delimiter, space, or comma, or something else) must be based on the language in which the names are being displayed, and perhaps the state of the UI displaying them. Perhaps variantNameForThePerson is not the constant quality of a resource it might appear to be, but instead a function of several things. Of course, if you really want to encode an order, another approach would be to use ordering structures in RDF. Those are not often very pleasant, but they work. In any event, your larger point about the difficulty of representing some things in RDF is well-taken.

> Thus like in the TEI scenario a bf:responsibilityStatement should allow internal structure (i.e. arbitrary XML) able to insulate and annotate forms of names, and furthermore should link this name to the person referenced by the name -preferably to resource-specific statements about this person.

I'm not sure I follow this. If the claim is that the sort of information in a responsibility statement is too complex for it to be obvious to us right now how it should be done in Bibframe, why would the solution be to include arbitrary complexity in Bibframe data? If responsibility statements are complex, can't we treat with them as entities in their own right? Wouldn't it be better to keep the Bibframe data simple and lightweight and let bf:responsibilityStatement be a link to some other kind of representation? For example, I might digitize a photograph of a copyright notice page, you might work up a stanza of TEI, and someone else might do something entirely different. If I understand the purpose of the responsibility statement correctly (and I welcome correction) it is to record the parties responsible for the resource described based on evidence from that resource.

An objection might be that in this scenario, it is only possible to do very little processing on the resources linked by bf:responsibilityStatement. I accept that, but I'm not sure to what extent it is necessary to process them. I mean that literally-- I accept that it is necessary to process them, but I'm not sure how much, and I don't understand how to answer that question without discussing what functions this information is meant to supply to patrons. This is why I think a point made by Karen Coyle and others is so very important: translating MARC (or RDA) structures into RDF is not going to work here. It's necessary to return at least to the underlying "intention and meaning behind the data element". As I wrote in an earlier message, better even to return all the way back to ask what purpose the data are meant to serve for a patron.

> 1. Our evidence fragment "Th. Mann" resides on manifestation level,  but creators and contributors are work- or expression  level elements: Therefore we cannot even think of  embedding the bf:creator statement within the arbitrary XML within the bf:responsibilityStatement

Bibframe makes bf:creator available for instances (sensibly, it seems to me). Or is the problem to which you are referring the lack of a class corresponding specifically to the notion of a manifestation?