[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Response to Laura, Wally, Brandon, Chris

----- Original Message -----
From: <WFreitag@aol.com>
> Yes, you're exactly right about what I'm trying to do and about the main
> problem I face in doing so. My approach at a solution is to make the
> chronologically noncontiguous. Each piece contains plot events or rules
> generating plot events, separated by gaps. The gaps have tags that say
> "insert such and such type of subplot here, with the following role
> assignments...". The pieces are not strung together in a chain, but rather
> stuffed inside each other. A "major plot" piece has the same amount of
> content as a "minor plot" piece; it just gets more layers of subordinate
> pieces stuffed into its gaps. As you surmised, each piece defines the po
> ssible courses of a specific conflict or theme.


I'd be interested to give you feedback on your approach; is there a way to
read up on your work, e.g., papers or essays?  How far do you plan to take
this -- from building parts of the architecture and publishing papers, to
building a playable (text, graphic?) system?

> My current opinion is that a new type of data-rich prose structure (think
> it metaphorically as the logarithm of a story, if that's not too alien to
> grasp), developed in concert with computational methods for manipulating
> structure, is the most promising approach. The structure may turn out to
> the conflict-centered plot pieces I described above, but there are some
> possibilities too. This approach also meshes pretty well with Jason's
> prescription: "We will build systems that empower individual humans as
> story-tellers. We will wrap them in increasingly clever algorithms..."

This sounds promising, I'd like to hear more about it.  I'd like to hear
your thoughts on how our approach is similar or different to yours.

Our data vs. process breakdown for the Facade project is something like the
following.  (Btw, we plan to write up more detail on this within the next
year).  As authors, on paper, we design a non-linear story.  We break the
story down into beats (a beat is typically a pair of lines of dialog, during
which a story value changes); represent beats in the architecture as some
sort of data structure; annotate these beats with preconditions, effects,
other interesting characteristics; write non-trivially sophisticated
behaviors (reactive planning code, animation, voice dialog) that can perform
these beats; organize the beats into "beat clusters", akin to scenes, with
their own particular annotations; write AI that can sequence these beats and
beat-clusters together, operating in reaction to the user's interactivity
and the general direction (arc) that we want story values to change in.  For
this to result in anything interesting and playable, the human authors must
create many, many beats for the system to work with, to allow for the
greatest degree of agency for the player as possible.  (In other words, it's
a TON of work.)  However, if the beats are cleverly designed the AI can
re-use them in a variety of different places, reducing the amount of
required authoring.  The system can potentially play beats and beat-clusters
in different orders, leave things out, etc.

For our purposes, we probably don't need to represent the abstract concept
of "conflict" explicitly; the conflict is engrained within the particular
beats and behaviors we author.  Therefore with this approach, brand new
story (ie, new lines of dialog, new behavior) perhaps cannot be generated,
but we feel there is so much to learn by first building a system that can
perform an interactive drama with even the aforementioned degree of
flexibility, that our approach makes sense for us.  Who knows, along the way
we may find opportunities for adding in more generativity that we haven't
have anticipated.

Part of the hand-waving magic in the above description is the first
sentence, "we design a non-linear story".  I feel this is at least half the
battle -- coming up with a way to design a story so that it can be
interactive in the first place.  We've found this fundamentally affects the
types and structures of stories available to us authors.  On a general
level, I think it will be easier to build interactive stories that are more
character-centric than plot-centric.  (Think of looser, freer stories like
Eric Rohmer films, versus intricately crafted stories like Casablanca.)
These can still be dramatic, just not as "perfect" as the best
non-interactive stories.  There is much to learn from the HTF/IF communities
on this point.

Note that at first, we were even more ambitious, we wanted to build a system
that could generate novel stories based on initial conditions that the
player established.  But we once we began discussing implementation, we were
forced to scale back some.  Even scaled back, it is 3 years of work for 2
full-time people (including designing & writing, coding, animating,