Newsgroups: rec.arts.int-fiction
Path: gmd.de!xlink.net!howland.reston.ans.net!gatech!concert!decwrl!world!dmorin
From: dmorin@world.std.com (Duane D Morin)
Subject: Re: Foreign Languages in IF
Message-ID: <CLwu0v.56p@world.std.com>
Organization: The World Public Access UNIX, Brookline, MA
References: <1994Feb25.140622.3424@schunix.dmc.com> <1994Feb27.160202.20730@oxvax>
Date: Mon, 28 Feb 1994 01:12:30 GMT
Lines: 64

In article <1994Feb27.160202.20730@oxvax>,
Jocelyn Paine <popx@vax.oxford.ac.uk> wrote:
>Don't reinvent the wheel! 

:)  I seem to be known on rec.games.programmer for doing just that.  Although
I like to think of it as "Learning about a subject enough to do it well, 
instead of trusting that someone else has already done it perfectly."

>True, there are still problems. One is the lexicon - most syntax-driven
>parsers will require you to provide morphological, syntactic and
>semantic descriptions for every word in your language. But if your dwarf
>spoke Loglan, you'd have to give all the players a Loglan word-book,
>since it is (to say the least) not widely known. So instead, why not
>give them a book specifying the English words they're allowed to use in
>conversing with the dwarf, and limit the size of your English lexicon
>that way?
>
>Another problem is trying to cope with ungrammatical sentences - and
>with sentences that, while grammatical to an English speaker, are not in
>the range covered by your parser's grammar. Loglan actually gives you
>some help here because it's possible to infer the syntactic class of any
>word from its consonant-vowel pattern. But again, if your dwarf spoke
>Loglan, you'd have to give your players a description of how to
>construct grammatical Loglan sentences. So instead, why not give them a
>description of which _English_ sentence structures will be accepted by
>your parser?

In both of your suggestions, you have the game limiting the English that
the player can use to converse.  But can you justify this within the scope
of the game, convincingly?  The best you could really do, it would seem,
would be to have lots of "The dwarf looks at you, puzzled." type of 
responses when you say something wrong.  I won't/can't write in 
the docs of a game, "Ok, if you want to talk to an NPC, here's the way
to form a sentence:"  I mean, that's not natural language.  That's 
commanding a computer to do something.  Conversely, it would be easy
to justify that, of course this new race is speaking in a language that
you don't know.  It would be a mistake to use 100% of a different
language, like Loglan, but something based on that could probably
pass.

>Incidentally, Eliza _was_ based on keywords, and could easily be fooled
>in exactly the same way as your example!
>

My fault for a bad example.  I meant, "Conversational AI, such as a 
conversation between two parties where the computer party can talk on
a subject chosen by the human."  More like a Turing Test wanna be.  Or,
Doug Lenat's Cyc project.  Eliza is just the most common example associated
with that type of AI.

>Jocelyn Paine


Maybe a simpler version of my original proposal.  Have the player have to
learn enough "dwarf" speak so that he/she'd be able to dress up like a 
dwarf and sneak past the guard into the gold mines by giving the secret
password, or something like that.  Player could earlier have hidden and
heard a conversation between two dwarves, and have a little phrase book
to figure out what it was that they said.  This doesn't satisfy my 
initial request, though, but it seems much more doable.  And it could add
some flavor.

Duane

