Tuesday, May 10, 2005

Chris Dent has posted a Purple response to Adina Levin's questions on Purple Numbers.

It's a good response, and I think the point about granularity and addressability is spot on.

Every time we manage to pulverise writing down to a finer granularity, to smaller pieces, even more loosely joined, exciting things seem to result.

I'd say the blog entry and wiki page represent two fine grained elements (relative to books and newspaper articles). Purple-somethings should, by analogy, be similarly interesting.

But I still disagree with the other intuition in his thought :

"The purple numbers are meant to be entirely meaningless identifiers (not labels or names) and therefore not present any information about the content they identify (no sense of time of creation or of sequence). This makes them portable and stable in the extreme."

Which seems to suggest that the "meaninglessness" is a good thing. That's the natural "computer science" way of thinking, of course, : let's give things unique, arbitrary labels, and bind the semantics later.

But it seems this is the thing which always limits the growth of systems. (In my slogan "Abstraction Doesn't Scale")

The problem is, that it's pretty expensive to make in-links to chunks of data with arbitrary names. I have to go look at the page and look up the number. Or I need a good UI to help me - one I have yet to see.

Contrast this with what's shockingly radical about wiki : when I want to make a link to another topic I just guess the ID for that topic. I know the page about purple numbers will be called PurpleNumbers. I don't have to look. And I don't need the software to do any work to help me find out. The knowledge is extant in my understanding of my language plus a couple of wiki-rules. The semantics of the label are bound by everyday practice.

As well as making link-authoring dirt-cheap, it also has other surprising effects. It allows page-making and link-making to be decoupled in time in both directions. Not only can I make the link without knowing if the page is there, I can make the link to a page which doesn't exist. And if someone creates it later, the link will work.

In any system where nodes have arbitrary labels, the chances I can guess the arbitrary nodeID before the node exists, and that a later author of a node will conveniently use my label, is pretty remote.

The people who are starting to understand this are the Folksonomy people. Tagging is another strategy where semantics are fixed by everyday language use.

So now I'm wondering. The individual tagging of paragraphs is a good thing. The "numbers" are the bad thing. What if the paragraphs could themselves have names (as with Wikipedia's sub-headings) based on standard, easily guessable, conventions. Or could even have several tags, which could be guessed?

Dent is hostile to this notion, complaining of labels posing (miserably) as identifiers.. But I'd say "labels posing as identifiers" is as good a definition of one of the greatest wiki virtues as one can imagine. Not to mention the core of folksonomy.

No comments: