Lists Home |
Date Index |
AI researchers I've talked to laugh out loud.
So much of the web hype ignores lessons learned.
It isn't that the Semantic Web, XML et al don't
work: it is that what is promised and touted
usually is beyond the technology and infrastructure
of a WWW-like system. Will it work? Eventually
if the slow pace doesn't kill it with boredom.
Lessons from Popsicle sticks:
Why did the boy stick a hose in his friend's ear?
He wanted to brainwash him.
Self-describing and extensible are two of the bigger
XML hypehose words. The VRML to X3D project has
been amazingly revealing. From the first to now
almost the last, one message has had to be posted
to the developers list: XML is just syntax. It
is stupefying how many VRMLers didn't get the implications
of that and not because they aren't bright well-trained
CS majors but because VRML97 was clever. The syntax
was lean and mean, the curly brackets looked enough
like the coding style they were familiar with, the
structure reflected a node system that was enough
like the OOPies they implemented that the whole
thing made elegant sense. The trick was, VRML isn't
a metalanguage. It has exactly one application.
The abstractions of the node types don't appear
in the syntax, just in the specification. So
the metalayer confusion has been horrific. That
is working out now, but it took a lot longer
than anyone would have guessed.
Other than interoperable tools, so far, XMLizing
VRML hasn't improved it much. We already could
take XSLT and output VRML97, so the metadescription
to rendering declaration pipeline worked. Using
an XML editor to create VRML without the ability
to pick on graphics objects improves nothing.
Over time, namespaces and XSLT templates do result
in a library of well-tested, well-understood
artifacts that really can improve the lot of the
VRML author, but for the VRML rendering implementor,
it's a slow slog through the trenches of a verbose
syntax with structural rules quite different from
the lean world of VRML (VRML has no root). What
we may get eventually are standard tools for
profiling but so far, namespaces don't play a
critical role in X3D.
A metalanguage with syntax unification as its
primary legacy exacts a price from the implementors.
SGMLers know this one. You don't get much from
the metalanguage itself; all the power comes from
the application language frameworks devised for it.
The so called complexity of XML comes not from
XML but from the XML systems, the interoperating
applications. This is dull news but it is what
makes the press yammer for standard DTDs and schemas,
makes the XML-Devers want to refactor the languages
and folks, that is business as normal.
From: dehora [mailto:email@example.com]
> From: Rick Jelliffe
> But I do agree that the term "self-describing" does seem open
> for misinterpretation by anyone who has not looked at XML for more
> than a minute: it may suggest that XML forces one to use names
> from some global controlled vocabulary, or that it does more than
> a simple sanity check on the names.
In the XML world, 'self describing' seems to mean, self-describing, ie,
XML data holds all the information needed to process and understand it.
The problem is not with human readers (heck, they'll interpret almost
anything), but that calling XML self-describing ascribes magical
properties to, and/or magically simplifies machine understanding of, the
data (presumably due to the data being tagged). Which is likely as not
to mislead and eventual disappoint many people. Certainly I'd like to
know what descriptive power XML has that would allows us to declare
victory in this regard and throw away all our domain specific procedural
code and business logic, replacing it with a general problem solver.
AI researchers must be kicking themselves ;).