Re: Question on the boundaries of content negotiation in the context of the Web of Data

I don't mean to defend CN or any particular theory of documents, but
here is how I've come to understand what Tim has said on the subject:

Think about one of the canonical use cases, bookmarking. Abstract
document (AD) = what you bookmarked = something satisfying the reason
you bookmarked it = something that says what you expected it to say.
That is, an AD is a set of expectations for its representations. I
think this means that different purposes implies different ADs and
therefore different resources. The different representations have to
make a reasonable effort to be similar, because a visitor (one who can
process all languages and media types in question) who bookmarked
based on seeing one representation has to be happy when encountering

A similar use case is you visit an AD in your browser, say "that's
cool", email the URI to your friends with the statement "this is
cool". You would be peeved if your friends went to look at it and what
they saw was really lame, right?

I don't think anyone would be happy if a .ttl were quietly substituted
for a .png or vice versa. I bet it's possible to come up with
particular .ttl/.png pairs where this might be possible (e.g. the .png
is a scan of the .ttl), but I don't think your example is one of them.

The idea seems to be inherently unformalizable, as it corresponds not
to a set of testable requirements but to whatever would induce
satisfaction among typical page re-visitors. I say "typical" because
different visitors may be unhappy about different things. E.g. I may
tell my friend to look at the top of page 7, and when my friend looks,
all they see is HTML without page breaks; I'll be peeved because for
me the PDF pagination had importance and I had no way to know CN would
swap what I saw for something else. On the other hand if I had said
look at section 4 I might not have cared about CN, if other
representations preserved section numbering.

Of course, if the URI owner has done nothing to set my expectations
(e.g. say whether or not page layout is part of the resource's
"identity"), I have no right to expect anything at all. [This is one
reason I'm so interested in metadata protocols and formats: there
needs to be a protocol and a language for communicating these
expectations.] But people form expectations nonetheless, and that's
why agreement is urged between representations.

A good practice note might be in order, since the temptation to use CN
in exotic and "lossy" ways that jeopardize bookmarking and casual URI
sharing seems to be persistent.

I'm as starved of citations on this subject as you are.


Received on Thursday, 12 February 2009 15:57:45 UTC