Sunday, 8 Jan 2006

En lieu of a coherent article on the thoughts therein, I present a lightly edited log of an IRC conversation that I kept in my drafts folder for the better part of a year. The subject of the surrounding discussion was initially wikis, then drifted to thoughts on implicit linking; at some point down the road, the following exchange ensued:

  1. JMP: One of the online encyclopedias just makes every word a link.
  2. AP: Interesting… You’d need to able to define multi-word links, though. I guess real hypertextuality with that approach requires that you be able to make words part of multiple links; e.g. in “flogging a dead horse”, “flogging”, “dead”, and “horse” would each be separate links, but the phrase as a whole would also be a link. And then I guess you need some indicator of how relevant the linked document is with regard to the one you’re currently reading; simply: “How relevant to the current page is the target of that link? How much does the stuff in this document have to do with the stuff in the linked one?”
  3. JMP: Except you really need context for words and even phrases. Expanding an acronym to something out of context is useless.
  4. AP: No, that’s the point: any point can link to multiple things, so a single acronym can point to multiple expansions.
  5. JMP: Listing all possible expansions of CSS in use in every field of endeavour isn’t useful either.
  6. AP: Sure, but the relevance indication would fix that: if you’re reading a page on HTML, the link to the Cascasing Stylesheet expansion would be indicated as much more relevant than the one to Content Scrambling System; in a page on DVD it would be vice versa. What strikes me as really beautiful about this is the lack of explicit linking. The things we currently make explicit links would naturally be those parts that are most relevant to the document in question. Of course the number of possible word sequence links in a single document alone is astronomical, and then you have to generate a relevance rating for all of these relating to the document…
  7. AP: Do you see what I’m talking about, though?
  8. JMP: Similar to the “semantic web” concept that’s been around since the ’50s.
  9. AP: Yes, and actually close to the way the brain works… everything is interlinked, and some links strengthen over time… You could read any piece of text as directedly or meanderingly as you want: look for specific info; or just go freewheeling around; or any shade of variation inbetween…
  10. JMP: And set a minimum level of relevance.
  11. AP: Actually, no: imagine relevance is indicated by colour in some fashion – intensity or some such, i.e. assuming a bright background, very relevant = bold dark red, nothing to do with this = light gray –, and then you wouldn’t have a threshold, you’d set a gamma curve. It doesn’t actually have to be implemented in terms of colour, that’s just the principle. That idea would even allow you to specifically go on free-associating tours by requesting that the strongest indication be given to only mildly relevant links.
  12. AP: It might be very interesting (but even less feasible) to start thinking about following the links on those pages, too – like weighing the raw relevance of a page by a metric derived by the accumulated relevances of the pages most relevant to that page. But that search space is gargantuan.