[
Lists Home |
Date Index |
Thread Index
]
[Mike Champion]
> 8/19/2002 7:06:40 PM, "Thomas B. Passin" <tpassin@comcast.net> wrote:
>
> >
> >Well, you do want to remember that we have both computers and people
> >involved. For people only, we want nice readable names and can make a
lot
> >out of a little context - plus we understand about furniture when we see
> >"chair", etc. For a computer, you just about need CycL to do anything
> >human-like with "chair", absent a schema-like something or other.
>
> I think this gets to the heart of Simon's point: He's asserting, and
> I'm agreeing, that you DON'T need something like Cyc or a huge
> RDF ontology to disambuguate / figure out how to process markup
> via its context rather than an elaborate system of identifiers.
> You probably won't get the accuracy with a pattern matching
> approach as you do with an identity-determination approach, but
> you may well hit an 80/20 point in actual costs/benefits.
> Elliotte Rusty Harold seems to have made a similar point in the
> "generic xml" thread on the TAG list, and called down the wrath
> of various WAI people for his pains -- you may need strong AI
> to recognize what is a "headline" in a loose XML+CSS system
> rather than a well-known standard, but you can probably make
> a very good guess with some pattern matching heuristics.
>
> XML lives in the middle ground between purely human-driven systems
> and purely machine-driven systems. Compromises are necessary --
> it's got to be somewhat human-authorable, and somewhat machine-
> processable, but if you go to far in either direction you miss
> the point. If ther are machines at both ends, you might as well
> use ASN.1 protocols; if there are humans at both ends you might
> as well use PDF or HTML. The point I take from this is that
> if an architecture requires human authors to type long URIs to
> get an unambiguous identity, there are inevitably going to be
> errors that make all that logic moot (Recall the recent post
> about a colleague practically going postal when he discovered that the
> bug he had spent days tracking down was due to using "w3c.org" rather
> than "w3.org" in a namespace URI). In other words, identity-based
> pay for their accuracy and machine-friendliness with fragility if
> the identifiers get screwed up somehow.
>
> Finding the right balance between *easily* machine processable
> markup and *easily* human authorable markup is not trivial; I
> think all Simon's trying to say is to remember the human element,
> both as a part of the system you have to work with and as a
> metaphor for how data can be processed using patterns rather
> than formal identities to associate markup with processes.
>
Actually, I think we agree on just about everything except perhaps how
possible it would be to have the computer end figure things out from
context, which I still see as fairly hard. And I am definitely aligned with
bringing the balance back more to the human side.
Hmm, heuristics giving a less than perfect accuracy - it is starting to
sound like quantum computing where the result is a probabilistic sum of
amplitudes rather than a single logically entailed result. Heuristic
quantum markup, I like it. Come to think of it, the Dirac braket notation
is nearly markup already (e.g., <a>=<m|A>)
Cheers,
Tom P
|
