[
Lists Home |
Date Index |
Thread Index
]
> From: Joshua Allen [mailto:joshuaa@microsoft.com]
>
> First, you are completely violating Axioms 1 and 2a of web
> architecture, by making the identification function of a URI
> be dependent on contextual information.
If axiom 1 was the case:
1: RDF would not need a Model Theory. We'd have no requirement for
assigning interpretations to URI vocabularies. But, oddly enough that is
not the case; we do have the requirement for a denotational semantics
for RDF and a number of people have spent a lot time on it.
2: we could throw away 2a as redundant. Note: RDF MT requires this
one.
The axioms are desirable, but not observable. And perhaps 2a is not
desirable; this would suggest the languages of the semantic web can't
evolve and will continually need to be replaced. The only computer
language I know of that comes remotely close is Lisp and this is
precisely because you can change what the symbols mean by writing
macros. I have no idea why they're called axioms. The web is clearly
not a formal system.
> Second, you haven't escaped the problem at all. You are
> still defining your words in terms of other words. You still
> need to achieve global consensus on *some* set of globally
> recognized words. How do you achieve consensus on what
> TimsProperties:Is is? For your URIs to be useful globally,
> they would need to be hoisted or bootstrapped out of your
> local context with at least a core set of globally
> unambiguous terms. But what would you call *those* terms?
> Global Resource Identifiers? I am sorry to say it, but you
> can't have a conversation unless you have words. And words
> need to exist *before* dictionaries can exist.
People converse all the time with loosely defined words. People have
different models of the world. CS people have different models of data
and communicate. Just look in a dictionary; there are more definitions
than words. How would this not be the case?
There are more resources than identifiers. Ambiguity cannot be designed
out, unless we are prepared to radically restrict what people are
allowed to say. Ambiguity is mathematically inevitable when we want to
speak using a countably infinite set of identifiers. Thus. A many to one
relationship between identifiers and resources. Be able to say anything
about anything. Choose one. I prefer the latter since such a
relationship is not inherent to information.
>It is theoretically possible, but it
> would have destroyed adoption, opened the door for vendors to
> profit from lock-in, and would not scale. The same with the
> semantic web. If you require something as basic as *words*
> to endure context switches which have drastic impact on
> meaning, you might as well not have words at all.
What you're asking for is far too restrictive.
> However, IF I want people to be able to globally understand
> me, I will try to use words that have more globally accepted
> meaning. YOU can do differently if you want. Just don't
> expect many people to understand. And for God's sake PLEASE
> don't run around giving people advice about how to choose
> words for maximum audience comprehension.
I hope you're not going to start accusing people of being stupid and not
sane if they do this (I read your TAG post).
> IMO, it would be far more useful to publish some guidance which says:
>
> "If two people independently use the same URI, absent of any
> other contextual information, they should be able to have a
> reasonable degree of confidence that they are identifying the
> same resource. URI naming schemes should be selected
> carefully, and people are strongly discouraged from naming
> resources in a manner that works against this confidence."
Nice idea, but we have no architecture for probabilistic reasoning. My
guess is we'll build one.
Bill de hÓra
..
Propylon
www.propylon.com
|
