[MUD-Dev] re: Language Parsing for NPCs

Robert Zubek rob at cs.northwestern.edu
Tue Jun 26 13:44:20 CEST 2001


Bruce writes:

> So, handling the processing and recognition of those actions
> needn't involve language parsing at all.

yes! that's what i meant by muds placing interesting constraints -
they make perception *really* easy. unlike real-life AI systems such
as robots, which have to do monstrous amounts of processing on their
perceptory inputs (and usually get it wrong anyway), game bots have
*perfect* perception. i mean, they can even have complete
omniscience if the game designer wants that. :)

but indeed, just seeing the world in perfect detail doesn't mean the
bot will be able to make any sense of it. most of human interactions
require rich background knowledge to be understandable, and this is
where the big AI problems begin - because even though the bot can
access all of the data structures in the game, it still doesn't know
what the players are doing on a personal and social level.

i mean, even considering a made-up interaction such as this one:

   X says: "i can give you 30gp for that sword"

   Y throws an amused glance at X

even if we do parse it perfectly, and are able to extract the
actions being performed (ie. x speaking, y glancing amusedly), and
the actions declared to be performed in the future (exchanging some
particular sword for 30gp), that still isn't telling us enough. what
the bot really needs to understand is the high-level action: oh, X
is trying to buy a sword from Y!

> Regarding the rest, have you read _Le Ton beau de Marot_ by
> Hofstadter[1]?  Judging from what I've seen of player or character
> discussions on muds, I'm not sure that I'd agree that the
> environment is limited sufficiently to simplify the problem.

i'm not familiar with that book, but from the amazon reviews i'm
guessing it's about the difficulties of language, especially in
terms of translation and cognition?

well, i'll admit readily, there is simply no way that, in any
foreseeable future, we will be able to do language processing good
enough to *understand* most, or even much, of what the players are
saying.

but chances are, we don't need to. 

consider how a recent immigrant might cope in a new society. even
when someone's vocabulary is really limited, and they don't
understand what people around them are saying, they can still cope
pretty well when interacting with others - they make up for absent
language skills by interpreting the actions of those around them,
and fitting them into the model of how they think people ought to
behave in different situations.

which offers a glimmer of hope that our present, simple language
processing systems, combined with a robust representation of what
the players are doing and what situations they're in, may be just
enough for the bot to very roughly understand what's going on around
it - their interactions would be far from fluent, but anything would
be better than the status quo. :)


--
Robert Zubek 
rob at cs.northwestern.edu
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev



More information about the mud-dev-archive mailing list