Notes on Conneg for TAG f2f

Henry S. Thompson
2 March 2009

1. Conneg.

1.1. Review of the thread

Hausenblas asks

"is the PNG *representation* derived via conneg from the generic resource <> equivalent to the RDF in Turtle?"

Two possible answers:

  1. Yes, because you own the resource and by doing the conneg that way you are asserting that they are equivalent for your purposes
  2. But you are being unhelpful in asserting that equivalence, as it is unhelpful to users. The png response implies the resource is an IR, i.e. that the generic URI identifies "an image of a house". The RDF response (modulo the lack of redirection) implies the resource is not an IR, i.e. that the generic URI identifies a house.

The overloading of conneg to attempt UAM is bad for web architecture, but the correct response is to provide a clean UAM story, which should relieve the pressure on conneg.

WebArch itself talks mostly about 'consistency', including the intriguing line "Improper use of content negotiation can lead to inconsistent representations." C.f. JAR's 0170, I think webarch's comments on CN are all contextualised by an assumption that we're talking about png vs. jpg or 1200x1600 vs 600x800.

Xiaoshu's 0105 is interesting, in that it raises well the issue of what the resources are that are named by the URIs returned as Content-locations: in responses to conneg. Do we have an answer to this one? In Raman's alternatives discovery finding? Hmmm, yes, he talks about 'generic resources' and 'specific resources' as if they were distinct. . . Language conneg, UA-type conneg (VARY?) and Content-type conneg are possibly different in this regard?

My top-of-my-head response is that all three URIs identify the same resource. I think we're forced to that conclusion by the fact that the owner asserts that both representations are representations of the resource identified by the original URI. . . Can the current state of AWWSW prove this conclusion?

JAR's narrative in his 0172 makes me uncomfortable, and what it is that makes me uncomfortable is precisely the possibility of two things which are so different as PNG and RDF being served as alternatives. . . I am inclined to reject even the "PNG of the graph" or "graph of the pixels" 2nd-order thought experiments, because what's crucial wrt Media Types is that applications which understand that media type will produce consistent effects. In none of the three cases is that true.

Last-minute thread

Axiom 1 is true with or without frag-ids. . .

I don't think Axiom 2 is correct. . .

1.2. Some thoughts of my own

Conneg only appies to IRs [i.e. you must serve them all with 200, and if you lost one, you'd serve it with 200][there might be a case in which you serve them all with 303---didn't we look at that in Cool URIs for the SemWeb?]

One possible exercise would be to enumerate application types (as close to media types as possible, but allowing for e.g. application/xml+svg to be in with image/jpg) and make assertions by cases. Here's a straw man: with certain listed exceptions (e.g. svg, application/xml plus a stylesheet) conneg SHOULD involve the same primary media type.

Here's a completely different straw man: conneg should only ever involve two Manifestations of the same Expression.

Note that

"Translations from one language to another, musical transcriptions and arrangements, and dubbed or subtitled versions of a film are also considered simply as different expressions of the same original work. [emphasis in the original]"
(section 3.2.1, Functional Requirements for Bibliographic Records, IFLA, 1998, file:///e:/gregory/d/My Documents/FRBR/

Score and performance are different expressions!

"w1 Franz Schubert's Trout quintet e1 the composer's score e2 a performance by the Amadeus Quartet and Hephzibah Menuhin on piano e3 a performance by the Cleveland Quartet and Yo-Yo Ma on the cello" (ibid., 3.2.2)
"The boundaries between one manifestation and another are drawn on the basis of both intellectual content and physical form."
"Changes in physical form include changes affecting display characteristics (e.g., a change in typeface, size of font, page layout, etc.),. . . "

Photographs and paintings are works [for paintings I guess expression and manifestation are typically 1-1]

In FRBR, attributes are associated with specific entities, and one of the attributes of manifestation include "File Characteristics (Electronic Resource) File characteristics for an electronic resource include standards or schemes used to encode the file. . .and other characteristics that have a bearing on how the file can be processed."

Which of course brings me back to my favourite candidate for a 'definition' of IR: It's an expression. That is, an IR is anything which a) fits within the FRBR ontology and b) qualifies as an expression. (Works are not IRs, they're abstract; whether manifestations are IRs depends on where we go on the 'specific resource' issue).