[MUD-Dev] Parlez vous NPC?

[Matt Chatterley]

On Sat, 21 Mar 1998, Chris Gray wrote:

> [Matt Chatterley:]
> 
> Nice analysis!

Thanks!
 
> :The weather is bad. What do I think?
> :
> :Which leaves us with the big question. How does the NPC respond?
> 
> Perhaps the response can be improved by going back to the original input
> and saving a bit more information about it. The NPC knows that the player
> thinks the weather is bad because of snow, specifically. So, the NPC could
> search through its learning for anything about snow. If it finds that
> snow is linked to cold, it might be able to respond:
> 
>     Yeah, I really don't like the cold.
> 
> or, perhaps it knows that snow is slippery, so:
> 
>     It's the slipperiness that is a problem for me.

Yeah! This is why I'm really starting to like the idea of giving it 'word
association' data rather than lists or combinations of words (a dictionary
of sorts), and also allowing it to make its own links.
 
> Output might come via a set of canned outputs that have slots that can
> be filled in. The larger the set of outputs, the better the effect.
> Also, the sets of canned outputs could be keyed to information that has
> been gleaned from the already-done analysis. So, in this case, the set
> of relevant outputs would perhaps be those where the NPC wants to
> commiserate with the player with regards to a specific topic.

This is the system we currently have in effect - there are a handful of
canned outputs so to speak, for each topic (over a scale of 1-5 for
'tone'), with a few tokens that the NPC fills out. I'm going to try out
dynamically buildings sentences when none of these canned outputs matches
(or on a random chance), because the sheer volume of canned outputs that
would have to be written is scary. :)

--
Regards,
	-Matt Chatterley
Spod: http://user.super.net.uk/~neddy/spod/spod.html