Accumulated various topics
Jorn Barger
("Perdition's Flames" you say, Mike? Funny coincidence, there... ;^)
Last week I made some first moves towards building 'jerks' out of
miscellaneous fragments of *poetry*.
But some of you are still asking, I expect, ***why poetry?***
I'd like to argue for a new paradigm for cognitive science, that starts by
collating literary descriptions of human behavior, and moves on to computer
simulations, for now *exactly like TADS*. And the laboratory/ measurements
part (that is usually considered the startingpoint for 'real' science) need
only come into it after you have a stable model of behavior that holds up
beyond the easiest challenges...
And people like Roger Schank are going to have to be left behind, because, as
promising as their viewpoints may have seemed 20 years ago, they're fixated on
pseudo-macho pseudo-scientistic ideals, so that poetry and adventure games are
*sociologically taboo* to them. (Dave Baggett, how do folks at MIT feel about
gaming? Do you get due respect?) In my case, anyway, Schank has used his
manipulative machiavellian intimidation strategies to lock good science out,
for now, so... fuck him! History will be in our corner...
But I favor *poetry* over any sort of prose *solely* because it's fun to
reread, and precise and vivid. And in the romance domain, poetry anthologies
have been *pre-sorted* for domain-relevance, more thoroughly than any other
resource...
So, in the process of winnowing and rewinnowing quotes that seemed potentially
useful, the hot-warm-cool-cold dimension 'jumped out' at me, along with a
*misexecution* dimension I'm still pondering ("You're so excited you knock
over your drink...").
(First topic shift:)
I want to restate my interactive-Moby-Dick argument, because it had nothing to
do with bodyparts, and may have gotten under-looked in all the petit-porn
hullaballoo back there...:
Advances in storytelling AI *will* gradually allow more and more detailed and
realistic levels of interactivity in our fictions.
But this doesn't mean that great interactive literature is impossible with
current technology-- if your prose descriptions are artistically written, then
your IF will be literature, even if it's hardly interactive at all.
So I'm arguing that WordPerfect supports 'zeroth-order' interactive
literature, assuming you've loaded in a textfile of great fiction like Moby
Dick....:
"Ahab cries, 'Ha! Starbuck! but the deed is done! Yon ratifying sun now waits
to sit upon it. Drink, ye harpooners! drink and swear, ye men that man the
deathful whaleboat's bow-- Death to Moby Dick! God hunt us all, if we do not
hunt Moby Dick to his death!'"
> CONVINCE CREWMATES TO MUTINY
I don't know how to do that.
> TELL QUEEQUEG JAIL AHAB
I don't know how to do that.
> YELL "AHAB IS A JERK"
I don't know how to do that.
> WHACK AHAB WITH STICK
I don't know how to do that.
> DRINK CUP
I don't know how to do that.
> GO AFT
I don't know how to do that.
> SOUTH
I don't know how to do that.
> NEXT PARAGRAPH
"The long, barbed steel goblets are lifted; and to cries and maledictions
against the white whale, the spirits are simultaneously quaffed down with a
hiss..."
(Shifting gears yet again:)
Here's Mike's own survey of the *range* of current TADS syntax:
go north north n up
take the box
put the floppy disk into the box
close box look at disk take disk out of box
look in box
wear the conical hat take off hat
turn on the lantern
light match light candle with match
ring bell pour water into bucket
push button turn knob
eat cookie drink milk
throw knife at thief kill troll with sword
read newspaper look through window unlock door with key
tie the rope to the hook climb up the ladder turn the knob
jump
type "hello" on the keyboard type 1234 on the keypad
get in the car get out of the car get on the horse
give wand to wizard ask wizard about wand
robot, go north. push button. go south.
take the box, the floppy disk, and the rope
put disk and rope in box drop box and ball
take all ['all' is specially coded for take and put, it seems]
put all except disk and rope into box
take everything out of the box take all off shelf
take the box open it take the disk and the rope
put them in the box take the disk and put it in the box
take box. open it.
unlock the door with the key. open it, and then go north
Besides bodyparts, what's missing here are plurals and possessives. Bodyparts
would best be implemented to allow any possessive construction like:
kiss jerk's nose
kiss his nose
kiss nose of jerk
And parts of parts would fit fine if we could allow for recursion:
bite finger of hand of jerk (ow!)
It would be nice if you could instantiate 100 jerks without having to
instantiate each bodypart for each of them. So long as bodyparts don't need
individual properties (clean, dry), can't all jerks share one set of
bodyparts, solely for the *methods* they provide?
Mike, what are your plans for 3.0? Neil Guy says he's hacked plurals, as well
as feet and hands and clothing-layers. I-for-one would like a look at his
code, despite his modest disclaimers about its inelegance... ;^)
I'm thinking even in TADS 2 we might do bodyparts as *specializations of the
actor*.
(Another topic-shift:)
I'm barrelling (ha!) towards proto-jerks, which don't really need these
newfangled details: the first demo has gotta be titled "Jerks on the Loose!"
Anybody want to contribute some pickup lines, or droning monologs?
(Someone suggested I advertise on alt.romance or rec.arts.books for some
better-informed 'consultants'... Anybody have a problem with that? I worry
about getting into sexism flamewars, eg.)
A hypothetical scenario, with a few slightly-subtle twists:
A badly-dressed guy appears at your elbow, breathing funny.
"Come here often?" he asks, dully.
> COOL
"Mmmnn." you say ambiguously.
"You a sports fan?" he enthuses.
> COLD
"No way," you say. [all banality-bank conversations could start with a yes/no
question]
"Did you catch Letterman last night?" he inquires, trying a new tack.
> COOL
You say nothing, but perhaps betray a glimmer of interest.
"My favorite thing on Letterman was always the Angry Guy..." he rattles on.
> COLD
You ignore this remark completely.
"Yeah, well, nice talkin' to ya," he scowls, and slouches away.
You feel a twinge of remorse, but then remind yourself how creepy his
breathing was.
[Yeah, I can't help it... I'm a whatever-the-opposite-of-misogynist-is.
I hope no one is *too* alienated by this.]
David Baggett
>And people like Roger Schank are going to have to be left behind,
>because, as promising as their viewpoints may have seemed 20 years ago,
>they're fixated on pseudo-macho pseudo-scientistic ideals, so that
>poetry and adventure games are *sociologically taboo* to them. (Dave
>Baggett, how do folks at MIT feel about gaming? Do you get due
>respect?)
Game-oriented stuff is mainly the Media Lab's territory now, and hoopy
virtual-reality-ish pseudo-AI cyberspace things are more than welcome
over there. The people in the AI lab tend to think it's mostly crap,
however. :)
Perhaps unsurpisingly, the AI and Media labs aren't on the best of
terms now that funding is getting tighter. As a prominent member of
the faculty here said recently, "We [AI/LCS] think they're charlatans,
and they [Media] think we're pitiful nerds." :|
I guess I should also point out that IF is really only a hobby for me,
largely unrealted to my academic pursuits. For my research, I'm mainly
concentrating on computer models of phonology rather than, as is more
common in natural language work (and more related to IF), focusing on
issues of syntax. So I pay a lot more attention to the linguistic
community than I do to "traditional" AI work like Schank's. I don't
even touch issues of semantics (more on this later).
It is perhaps a bit of an ironic twist that most people here now
neither know nor care about Infocom-style games. I wonder what Stu
thinks of this ... being (at Thinking Machines) so close to what was
once the center of the modern IF universe, where there is now little
interest.
>Advances in storytelling AI *will* gradually allow more and more detailed and
>realistic levels of interactivity in our fictions.
Soap box response:
My view of IF is that it has reached a major plateau. In fact, that's
pretty much my opinion of AI in general. While many people in AI seem
to think that we're a field 40 years and tens of thousands of
researchers strong, I think the truth is more that we have wasted an
enormous amount of time creating (and making outrageous claims about)
superficial models of a very complex system -- our minds -- without
first understanding even its most basic, atomic elements.
How does thought work? How does learning work? Not necessarily
thought or learning in *people*, but in a mouse? In a newt?
Anything?!?!
We know plenty about the physiology of the brain, but once we get past
the "implementation details" we are in a complete wilderness. Right
now it's as if we're aliens who've never thought about computation, and
we've found a PC in the woods running Word Perfect (on batteries, of
course :), and we're trying to copy it just by looking at what happens
on the screen when we type things. We can make clever gadgets that
mimic the surface behavior in certain limited domains pretty well, but
without knowing how the darn thing really *works*, we're reaching a
point of diminishing returns. (If you think this is an unfairly absurd
analogy, you don't know enough about the miracles are minds our capable of.)
But this is often obscured in AI by the reckless speculation, heard
time and time again, that "we just need to take this [working] system
and generalize it a bit to get closer to solving [AI-complete] problem X."
I think the CYC project sort of epitomizes this misconception. But the
jury is still out on that, of course.
Another pet peeve: some people still talk about natural language like
researchers were 40 years ago -- just tweak the parser and "semantic
network" a bit and we'll have it solved! But language is much deeper
than that, in ways a layman would never imagine. Consider wh-question
formation in English:
1a John likes Mary.
1b Who does John like?
2a Bob thinks that John likes Mary.
2b Who does Bob think likes Mary?
3a Bob belives that Biff said that Alf heard that John likes Mary.
3b Who does Bob believe that Biff said that Alf heard that
John likes?
4a Bob believes the claim that John likes Mary.
4b *Who does Bob believe the claim that John likes?
(* means "ungrammatical")
You can't possibly account for this with a parser that you just "tweak"
-- you need a comprehensive theory of syntax to explain this. And this
is just one of many similarly complex phenomena, and only at one level
of the whole linguistic domain -- don't forget interactions between the
sounds and the morphemes! (BTW, 4b is grammatical in Italian. You'll
need to account for that too! :)
The point is that you can't get to the moon by climbing a tree. You
can't push any research program forever. At some point you need to
retreat to square one to make progress. I believe we're at that point
with all the issues that come to bear on substantially improving IF.
My feeling is that IF will only get better as a result of an intensive
program investigating learning in a variety of existing biological
systems. (I.e., bugs, mice, cats, etc.) Hence, I don't really worry
much about improving the "realism" in my games; instead, I focus on the
aspects that are interesting in traditional fiction: plot, setting,
characters, theme. The "interactive" part adds something very neat to
the fiction experience. But the more you try to fake, the more fake it
seems, and that's not, I think, going to change any time soon.
Next topic:
>Besides bodyparts, what's missing here are plurals and possessives. Bodyparts
>would best be implemented to allow any possessive construction like:
>[...]
>
>kiss jerk's nose
>kiss his nose
>kiss nose of jerk
You can already do this pretty easily. E.g., in UU2 you can refer to
your nose -- "examine my nose". But I wouldn't bother defining a
zillion body parts -- I think that kind of "realism" tends to only
confuse and distract the player. I still view these things as
*games*. They need to be fun and accessible. I thought that by not
paying attention to this aspect of game design, Dennis Cunningham
turned his otherwise excellent T-Zero into a very tedious game. Ditto
(from what I have heard of it) for Disch's Amnesia.
>And parts of parts would fit fine if we could allow for recursion:
>
>bite finger of hand of jerk (ow!)
I don't think you can nest PP's with TADS, unfortunately.
>Mike, what are your plans for 3.0? Neil Guy says he's hacked plurals, as well
>as feet and hands and clothing-layers. I-for-one would like a look at his
>code, despite his modest disclaimers about its inelegance... ;^)
Aiiiiie! Don't say TADS 3! After working so hard to port 2, I don't
think I could handle a major update in the next year or so.... 8P
>A badly-dressed guy appears at your elbow, breathing funny.
>"Come here often?" he asks, dully.
>
>> COOL
>
>"Mmmnn." you say ambiguously.
>"You a sports fan?" he enthuses.
This is a neat idea. I think this kind of thing is where IF can be
incrementally improved, if used sparingly. The only problem is that
it is a complete nightmare to implement. It's already time-consuming
enough to deal with all the object interactions -- here you would
have to hand-code hundreds of possible paths through the dialogue
for each character. (Yes, I am implicitly assuming that you will
have to essentially use table-lookup here, and that is somewhat
philosophical.)
>Yeah, I can't help it... I'm a whatever-the-opposite-of-misogynist-is.
It's probably telling that there's no term for this. ;|
Dave Baggett
__
d...@ai.mit.edu Natural Language Processing MIT AI Lab
ADVENTIONS: We make Kuul text adventures! Ask about Unnkulian 1, 2, 0, 1/2
PO Box 851 Columbia, MD 21044 USA / CIS: 76440,2671 / GEnie: ADVENTIONS