[Freddy]

Data suggested this had a thread for itself so here it is.

I like AIML, I think it's quite elegant and useful. People like SquareBear have shown what it can do. So I thought I would have a crack at AIML again. I was wanting to host it somewhere, so needed an interpreter. I like PHP and there was Program O, but I decided it would be interesting to write my own.

One thing I wanted to change about AIML was that an input has to exactly match a pattern to get a hit. I felt I could improve on that. So I went all fuzzy. This thread will deal with my ideas and what I have achieved. I'm not sure how much I will get done now the weather is so lovely, but I am working on the interpreter and it's preprocessing quite a lot lately. I have a few AIML tags working, but have a fair bit to still do. Anyway I will update as I go.

I'm working on spelling mistakes at the moment and still have a few things to try.


Just a little taster to show you how this fuzzy preprocessing fleshes out. Take this AIML pattern...

Code

<category>
<pattern>NAME THE ROMAN SOLDIER WHO IS SUPPOSED TO HAVE STABBED JESUS</pattern>
<template>Longinius</template>
</category>

Normally with AIML you would have to input the exact pattern to get a match -  but I can get a hit with lines like this :

"Who was that man who stabbed Jesus ?"

Making it a bit more flexible. Keywords are the key here.

In a different way I can get hits with spelling mistakes :

Code

<category>
<pattern>WHO WROTE CANDIDE</pattern>
<template>Voltaire</template>
</category>

Take note of the spelling of Candide. With my approach I get a hit with...

"Who wrote Candede ?"

Ask most AIML bots "What is a fish ?" and they will tell you...

"A cold-blooded vertebrate who lives in the ocean or fresh water."

Ask them "What is a fosh ?" and they will fall back to a default no-answer-for-that reply.

And similar things. One of the things I am doing I mentioned a while back here and that's using Levenshtein Distance.

Together with keywords and some other nifty ways of comparing strings it works quite well and makes my AIML bot a bit more flexible.

At the time of writing this I am still in the early stages. It's going to take me a while to see how effective it is. Sometimes I can see that a pattern match might not be desirable but as I have introduced a scoring mechanism maybe I can control that. The spelling mistakes I think will work well. It all remains to be seen.

[squarebear]

Nice work Freddy.

[Freddy]

Thanks. There's nothing like jumping in at the deep end 

[Data]

Thanks for this thread, so impressed with this whole idea, really, to my mind it has the potential to be quite a leap in AIML. looking forward to my first chat with it, maybe I wont have to worry so much about my awful spelling 

Keep at it, you will find your way out of the deep end. 

[Freddy]

Thanks Data, it's been interesting to work on so far

Some of my bits that I ordered to host it on arrived today. Got a few more things on the way. It would work as a really simple bot as it is, with just the features of AIML I have so far coded. I think I will go for full AIML compatibility now, well I will try...

[Bragi]

Sounds like an interesting approach Freddy.  You should get some nice results with that.