Architecture of theWorld WideWeb - Social Semantics: The Search for Meaning on the Web - page 33

The only architectural constraint that connects Web representations to resources
is that they are retrieved by the same URI. So one could imagine a resource with
a URI called http://www.example.org/Moon , that upon accessing using English as
the preferred language would provide a web-page with a picture of the moon,
and upon accessing with something other than English as the preferred language
would provide a picture of blue cheese. While this seems odd, this situation is
definitely possible. What binds Web representations to a resource? Is a resource
really just a random bag of Web representations? Remember that the answer is that
the Web representations should have the same content regardless of their particular
encoding if it is accessible from the same URI, where content is defined by an appeal
to Dretske's semantic theory of information (Dretske 1981). To recall, Dretske's
definition of semantic information, “a signal r carries the information that s is F
when the conditional probability of s 's being F ,given r (and k )is1(but,given k
alone, less than 1). k is the knowledge of the receiver” (Dretske 1981). We can then
consider the signal r to be a Web representation, with s being a resource and the
receiver being the user-agent. However, instead of some fact F about the resource,
we want an interpretation of the Web representation by different user-agents to be
to the same content. 10 From a purely normative viewpoint in terms of relevant IETF
and W3C standards, it is left to the owner to determine whether or not two Web
representations are equivalent and so can be hosted using content negotiation at
the same URI. The key to content negotiation is that the owner of a URI never
knows what the capabilities of the user-agent are, and therefore what natural and
formal languages are supported by it. This is analogous to what Dretske calls the
“knowledge” or k of the receiver (1981). The responsibility of the owner of a URI
should be, in order to share their resource by as many user-agents as possible, to
provide as many Web representations in a variety of formats as they believe are
reasonably necessary. So, the owner of the URI for a website about the Eiffel Tower
may wish to have a number of Web representations in a wide variety of languages
and formats. By failing to provide a Web representation in Spanish, they prevent
speakers of only Spanish from accessing their resource. Since the maintainer of
a resource cannot reasonably be expected to predict the capabilities of all possible
user-agents, the maintainer of the resource should try their best to communicate their
interpretation within their finite means. The reason URIs identify resources, and not
individual Web representations, is that Web representations are too ephemeral to
10 Of course, one cannot control the interpretations of yet unknown agents, so all sorts of absurdities
are possible in theory. As the interpretation of the same encoding can differ among agents, there is
a possibility that the owner of the URI http://www.example.org/Moon really thinks that for French
speakers a picture of blue cheese has the same sense as a picture of the Moon for English speakers,
even if users of the resource disagree. However, it should be remembered that the Web is a space
of communication, and that for communication to be successful over the Web using URIs, it is in
the interest of the owner of the resource to deploy Web representations that they believe the users
will share their interpretation of. So content negotiation between a picture of blue cheese and a
picture of the moon for a resource that depicts the Moon is, under normal circumstances, the Web
equivalent of insanity at worst, or bad manners at best.