Bot Contest

That file is PostScript and needs special software for viewing. I converted to PDF and posted it at http://newterminus.com/talk2.pdf For anyone interested. The font's a little weird, but it's legible.

That's some interesting reading, rex, thanks. However, the Forge is more advanced than the tricks he used, and I dont recall this guy winning the Loebner contest.

The article, to summarize, was written in 95 with a way to win the Loebner Prize. Clearly it failed, and I dont think it was such an amazing idea anyways. Basically they fellah just studied old Loebner transcripts and tried to make reactions for the types of things being said.

We've got 10 Personality Forge contestants in the contest so far!

To see 'em:
http://www.frontiernet.net/~wcowart/contest_links.html

I've just been going over the transcripts of the Turinng contest for the last couple of years and one thing stands out: the judges, almost without exception, are terrible typists, far worse than any of us meticulous folks...typos all over the place. This could affect key word recognition and response, couldn't it?

Yes, it sure could. I have a spelling module somewhere.. that can make guesses at the right word.. but it seems like a lot of work to put together..

Let's try it prof. Even if it screws up the site for a while (I'm sure it's not that easy, but it could probably really help).

I just looked into it and I'm going to have to custom make a spell-checker. And that looks to be a huge project. And even if it comes up with a large list of possible options, correctly choosing from them would be a wonder. Not that I cant do it, but I'll need an entire week set aside for it.

I guess you can't really do it then. An entire week!?! Imagine what could be done if you ever had en entire week!

It sounds pretty time-consuming. Even look at the big professional spell checkers out there. How often do they suggest the right word? I'd be a little worried about it's ability to multiply bad responses. It might totally wipe out xnones and xnonsenses too by always finding a word match, no matter what's typed.

My feelings as well Rex. Maybe the bot could just detect misspellings and ask something like... what do you mean by "senik"? And somehow put it all together in the next response (once the user corrects the spelling or provides a synonym). Once that gets going, we could probably build some kind of database of often misspelt words from the corrections. Would that be simpler prof?

Yes, you'll have to use metaphone, or a kind of soundex, no? And then compare to all the words available in WordNet?
Phew! What a work.
If you can make it in one week (usable on the site, I mean), you're a God Programmer, Prof

I wonder whether there are no PhD thesis on that subject.

At the Academy of Artificial Intelligence, could we not award Artificial Ph Ds?