Comment from Jamie Zawinski 1997 Dec 14

Send comments about this site to web+comment@lahosken.san-francisco.ca.us.

Jamie Zawinski sent in this comment.

My (Larry's) replies appear like this.

From: Jamie Zawinski
Date: 1997 Dec 14
Subj: Re: dadadodo

Lawrence Hosken wrote:

> You've made a wonderful thing.  Thank you.
Yours is nice too! Is that really unfiltered output, or did you do some selection in there? The sentences are awfully good.
>   Figure out the probability tables for depth-N and depth-N-1.
>   To figure out the next word:
>     flip a coin

That's interesting, I think I'll try that...

Hmm, the unfortunate thing about that is that it makes the *generation*
of the chain be probabilistic.  If you use the same text as input twice
in a row, you don't get the same histogram out.  I'm not sure that sits
well with me...

> This might break down if I started reading in huge amounts
> of text and N became large.  Then again, it might not.
I think that the larger the body of input text, the larger you want N to be, but it probably maxes out somewhere at (uh) 1/2 * length-of-a-typical-sentence or something like that. (Or maybe it's "verb phrase" and not sentence? I'll bet there's some language-specific constant about how many words our brains need in order to pull patterns out of groups of words. I'll bet it's not a very large number.)

I'm kinda cheating by including this email with comments that normal site visitors send me. Mr. Zawinski never would have seen my Markov page if I hadn't sent him some fawning mail about his. So I kinda solicited this mail. Anyhow.

[^]