Are random string really all that bad? For example, a seven character alphanumeric string would have 36^7 = about 78 billion combinations! A five character string would be 36^5 = 60 million. Surly a five character random string, e.g. k9rjf, would not be too difficult for people to remember? Put a year code on the front of it, and you have 60 million document spaces per year. For example, /2001/js90d/. If I say that "js90d" stands for my rants on URIs and WikiNames, then maybe people will begin to associate that concept to that random string? It is almost like inventing new words for a concept!!!
So, why not just invent new names for a concept? This is actually something I'd really enjoy doing because I like coming up with new words. Of course, this makes it a not quite random string, because it has to be pronouncable, but something like gumpo, or dirky (hmm... to realistic), or zylex is quite fun, no? It would be like giving your URI a name of its very own, and then assocating the concept with that name. I suppose the shortest any "invented" name could be is about five-six characters, and the longest should be about 7-8 for memorability. I'm not sure exactly what the statistics would be on that, but I guess you could create at least a million new and unique names. You could even search using Google to check for previous occurances. For example, I could name this document /2001/gumpo/ just for the sake of it. It should also be possible to come up with "random word generators"... in fact, I think that there are some already out there. In that case, you might be able to get your server to label your new document with a pronouncable name for you!
You'd have to be careful not to infringe upon any trademarks (check using a search engine), and you must be careful that it's not too good a word that someone else might think of in the near future. In other words, the idea isn't a perfect one, but it is a good one. Really, what we are doing is creating "memorable random strings", or MeRS (heh, heh). A series of related documents could share related names, for example they could all have the same first syllable. twipney, twipstle, twipt. Of course, the only problem with all of this is creating new words that don't sound completely rediculous. I suppose that new works mirror their owners, for example, all of the ones I come up with are rather odd and silly. Actually, having said "owners", it should be made clear that you shouldn't really try to make people adopt your new word for anything other than an identifier for your document, otherwise it becomes a "product name" rather than a "memorable random string". This is quite a big danger, and I'm not sure that there is anyway round it.
I suppose that this system works best for URIs that are "found" and used, rather than remembered for their content. It might be useful, for example, in a Wiki, except that a Wiki page is kind-of characterised by it's name. So maybe this isn't a useful idea at all? Well, I guess it would be in a "perfect" Web, where the only time you would want to see a URI is when referencing one in a plain text document etc.
A good idea for the syntax would be something with ambiguous pronunciation, e.g. dopynl. Combined with datestamps used as arbitrary strings rather than expiry indicators, this could prove to be a useful system.
Sean B. Palmer