Seasons

Or it could end up being schizophrenic.

Have you read the book "Godel, Escher, Bach" by Douglas Hofstadter?

An excellent and entertaining book. But have you read his Fluid Concepts and Creative Analogies - Computer Models of the Fundamental Mechanisms of Thought?
It's not as entertaining a read (in fact it's extremely hard going in places!) but it's a 'must-have' for anyone serious about programming their AI from scratch.

Check it out at http://www.amazon.com/gp/product/0465024750/

Then, statistics can measure how important or direct (how many times any one association is referred to) a link is,to give some law and order to the mass chaos of associations.

And you're going to program all these thousands (millions?) of statistical weightings by hand? I think you'd be better off with some sort of learning algorithms to adjust the association strengths as it runs. Indeed, Ally's response recycling is one such approach, but some broader processes that don't require user input at every step would be particularly useful - whether from transcript correction after the conversation, or the ability to feed preformatted conversations into her for analysis and deconstruction. I've always thought it's a shame Daisy and Billy can't eg: take a play, or a PF/IRC/MSN transcript and analyse the language use. It would be MUCH faster than having to spend hours having inane conversations while they learn.
Talking of Ally, do you want to announce her new home, or shall I?

I haven't read that one. I have a read part of Metamagical Themas.

I read that when it came out, but that's so long ago I don't remember it well now. I certainly enjoyed it at the time.
How I wish that he would publish another juicy book in the field. I've been waiting since Fluid Concepts came out (and that was ten years ago now I'm appalled to see, looking in the front of my copy!)
His only publications since then (bar a couple of essays) seem to be literary translations

Actually, ally does have the ability to read, analyze, and deconstruct text files (even I sometimes don't have the patience to sit there and teach her from scratch). the only drawback with that is the learning algorithm would miss out on the user's positive reinforcement, which should be done at an early stage before there is so much data that the chances of her saying something coherent go down.

And you're going to program all these thousands (millions?) of statistical weightings by hand?

that would be inpractical for a number of reasons, one being that any particular association would never be constant, they would always e changing as the entity grows. of course there would be a learning algorithm.

I tried out Watzer's sonnetplease, and I have to say, I am impressed. How is it done? Is there a way to detect if words rhyme? I did something like that with VB, detecting the last vowel in a word and using that and the rest of the word's consenants to make a simple rhyme detector, but how did you ever pull such a complex system off?

Announcing ALLY's homepage: http://www.be9.net/ally/index.htm

With thanks to psimagus for all your help

Crap! I still have school today! That's the second time this week we should've gotten a snowstorm...

Oh! It was easy, I just made keyphrases of nouns, verbs, adjectives, etc. that had the right inflection. For the plugins at the end of the line, I made ones that all had something with the proper inflection that rhymed with what I wanted them to. There's no system; the engine detects nothing. I just put it together. His sonnets are not complete yet, I plan to give them a bit more variety but that requires another set of plugins that are not all created yet.

interesting.. i'm trying to make a "poet machine" using a rhyme detector and AllY's sentence generation algorithm, but there are several problems to overcome, the main one being the ability to scale down sentences into something that sounds like a poem, but the algorithm works forward, adding words as it goes so it would be impossible to detect when it will chance upon a rhyming word.... looks like i'll need to reverse the process so that it finds a rhyming word and then works backwards- but then there's still the problem of finding the beginning of a sentence so that i dont end up with lines that begin out of nowhere... any ideas?

I don't think you can get recognizable rhyming poetry from an algorithm, and rhyme detection is all but impossible unless you program every word and what it rhymes with into the engine. You're going to have to do some very specific programming to get that to work, aren't you?
Though come to think of it, I also know some humans from whom you can't get recognizable rhyming poetry.