On Jan 13, 2004, at 7:26 PM, Julian Reschke wrote:
>>
>
> It was mis-specified (actually it wasn't specified at all, and as it
> wasn't UTF-8 it should have been).
OK, so the insistence on the encoding declaration being correct is the
"draconian" bit here. Thanks.
>
> I'll not comment on the rest because it seems to say that because of
> recent advances, we don't need a well-defined markup syntax. Somehow I
> doub this is true :-)
>
Not my argument. I'm saying that well defined markup syntax is
basically for machine-machine communication (although obviously the
übergeeks on this list can hand-author it), so machines are going to
be doing the work to produce it from human-authored slop. One still
needs good markup specs to define the template of the stuff that the
machine creates, and to allow the de-soupification to be done only once
in a processing pipeline.
The debate about Postel seems a bit pointless, since there's no way
that ordinary humans are going to be trained to be conservative in what
they produce and will insist on being liberal in what they consume. The
only alternative to despair seems to be to automate the drudgery,
ideally in the authoring tool, but more realistically in a downstream
filter. (Actually the RSS/Atom debate about this seems to be over
whose job it is to de-soupify, the syndicator or the aggregator).
There will be cases where one must insist that no dumb machine "fix"
the inputs, such as Tim Bray's example of the ill-formed stock
transaction message. I suspect there will be thousands of times more
cases, however, where it's more like the mismatch between the encoding
declaration and the character set in Sam Ruby's example, and machines
can be trusted to do the right thing.
A year ago, I probably would have disagreed, but I've seen how an
utterly stupid statistical tool (SpamBayes) has liberated me from spam
with a grand total of 1 known false positive (and that was a legitimate
message that sounded exactly like a spam, something like "the
information you requested is at such-and-such a URL") out of tens of
thousands of spams. Dave Raggett's tidy is another example of a fairly
dumb program fixing a lot of tag soup with minimal damage to actual
content structure. For that matter, Google and the next-generation
stuff such as Vivisimo do an awfully good job of making "judgements"
from tag soup, using inferred metadata rather than hand-authored
metadata. I don't think it requires strong AI to make a really good
guess at markup, especially in highly regular content such as weblogs
and news feeds.
