Phrontisterion 4

June 8-9, 2002

22 people attended. The weather was uncooperative; temperatures in the 50s prevailed all day Saturday and Sunday morning. On the week before and the week after Phrontisterion, high temperatures in the 80s were recorded. The gods were telling us something.

On Saturday, the group heard reports on various strategies for interactive storytelling; on Sunday, the group deliberated and prepared a joint report. Here are the group’s conclusions: (the comments of individuals are presented in italics)

1. Textual input systems
Command-line interfaces are undesirable as primary input systems, but are acceptable as hidden alternatives. Inverse parsers are desirable input systems for interactive storytelling. Systems in which all accessible words are directly visible are best; systems in which accessible words are available through menus are second best. When combined with speech recognition systems, inverse parsers become especially attractive. Ability to handle nested clauses is desirable. Inverse parsers should not require any particular word entry order.

TG: I believe that textual or any language based input systems are doomed to fail for 2 main reasons.
A. general users(mainstream public) will NOT learn a new language to be entertained.
B. general AI language processing is not even close to being solved. People have been trying to solve this for decades and continue to put their hand back on the hot stove. I have no doubt this problem will one day be solved but do not wish to hold my breath till then.

R: I read the spirit of the statement on Textual Input Systems as "Get the interface out of the way," and to this I add my assent. However: I’d like to temper the condemnation of the command-line interface (CLI) by pointing out that CLI offers a rich and arguably yet unmatched experience of Choice. This freedom is more apparent than actual, but this appearance of freedom is in itself highly desirable: there are no shortage of philosophers to remind us that all freedom is apparent rather than actual.
Another problem with systems that enumerate possible choices is that in making the possibilities explicit, they exclude the discovery of possibility. Lateral thinking puzzles are about the discovery of possibility, and will be missed in a system which does not facilitate them. A rich inverse parser will fare better in this respect than a menu-driven system.

2. Non-textual input systems
Adjusting dials for tonality expression as a supplement to other input systems is both viable and desirable. The problem of clash between the explicit content of the statement made and the tonality with which it is expressed can be resolved by several means. Dials can have limits imposed upon them by the nature of the explicit statement being made. Alternatively, the computer can respond to such a clash by asking, "Huh?" Lastly, there might be buttons available to indicate that the user is deliberately expressing irony, sarcasm, humor, etc.

Speech recognition technology is not yet adequate for use in interactive storytelling because it lacks tonality recognition and full language processing. However, when used in conjunction with an inverse parser, it is immediately beneficial.

Facial recognition technology has some interesting possibilities, but it is hampered by the demands it places on the user, possible problems with privacy, and insufficient emotional resolution. One "interesting possibility" is that the interactive storytelling engine might be able to dynamically alter its content based on the user’s facial expression feedback, employing facial recognition technology on an unconscious level rather than a conscious one and thus removing many of the demands placed on the user by a facial recognition engine requiring conscious expression control.

Speech generation is frigorific, but it will not be of general utility to interactive storytelling until it has the ability to express emotional tonality.

HHC: I disagree. Speech generation could be of general utility even before it has the ability to express emotional tonality, because even canned-sounding speech (in dialogue, at least) creates more emotional involvement in certain audiences than just reading text printed on the screen. Witness the popularity of poorly dubbed mass-market foreign films versus niche-market subtitled ones--not a perfect analogy, but I’d say it does demonstrate that the general public often prefers listening rather than reading in their entertainment choices. The other real power of speech generation is that it allows the size of the storytelling engine to increase without limit and become truly a storytelling (story-generating) tool rather than a collection of predefined stories. If, as in most if not all video games, voice actors record every single banal variation of every possible line, the number of possible lines ends up being severely limited based on the cost of recording voice talent and the size of the recorded voice-over files (a user shouldn’t have to deal with gig upon gig of voice-over data). However, with a speech generation engine, the number of different possible lines (and, implicitly, different plot twists and permuted situations) skyrockets because the cost of recording and storing the voice-overs for each line becomes irrelevant.

TG I don’t think time allowed us to explore this subject enough but I think a major point that was lost or not brought up was that this type of input is user limiting. This is a good thing in that the engine that receives this input can be much more robust and less likely to get into unsolvable problems.

3. The role of spatial computations in interactive storytelling
Spatial computations are desirable only as directly demanded by the dramatic situation.

TG: I was puzzled when this was brought up as a topic (Space Sucks?)and find myself even more confused by the conclusion. What are we saying here? This is only important when it’s important?

CC: Since this was my hobby horse, I shall break my editorial silence to explain. The games community, in my opinion, has taken the notion of spatial representation to such extremes that it does not even realize the obsessiveness of its dedication to spatial representation. Every game has a map; every game sports scads of spatial calculations. Drama doesn’t care about space; people appear and disappear in various locations without ever bothering to move between them. My original, undiplomatic thesis was that "Space Sucks"; the group watered this down to the more diplomatic observation that spatial computations are desirable only as directly demanded by the dramatic situation. I’ll translate that to: Don’t use spatial computations unless they truly meet some dramatic need of the story.

4. Simulation techniques:
Well-understood technology.
Appropriate for the medium.
Intrinsically interactive
Offers the possibility of emergent behavior
There are many development tools for simulation techniques

Lacks dramatic weight.
Dramatically unstructured.
The tools are not structured for interactive storytelling.
Simulation: sandbox vs directed play style drama
Easy to get lost in the details
Potential for the fetish of accuracy.
No dramatic themes.

TG: It would seem for the most part the disadvantages of simulations are areas that are potentially possible but not proven. My definition of a simulation is pretty broad and I have to wonder if we are talking about Graphical Language Design vs Textual Language Design?

5. Dramatic sublanguages
Potential for universality
Takes advantage of powerful human linguistic processing
Plausibly computable
Plausibly localizable
Well-matched to interactive storytelling because of its expressive richness
creole grammars would work well

If you rely on existing vocabulary, then its ambiguity will intrude into the sublanguage.
Tough trade-off between expressive richness and computability.
Unfamiliar phraseology
Difficult learning curve

TG: I believe players/users will not be willing to learn a new language to be entertained. I am aware that people have done so for text adventure games but my gut tells me those days are gone.

6. Character-driven approaches
We relate to characters
These are similar to simulations
These are similar to existing agent-based technology

No plot
No point of view
Requires very rich characters to work

TG: I believe you are ultimately going to need some form of plot to render an interesting story. Perhaps a plot could be divined from a characters needs/goals etc etc but without a plot you will end up with a simulation without a story.
7. methods based on the Aarne-Thompson index of folktale motifs
large existing source of information about stories

no empirical evidence of theoretical soundness
just doesn’t look workable
doesn’t seem easy to produce dialog

8. methods based on Vladimir Propp’s analysis of folktales
a good model for stories
more structure than Aarne-Thompson
usable as a high-level plot generation system

Propp’s expressions are too vague to be directly useful

9. The Erasmatron
scene based
includes moods, attributes, and relationships
the technology has been built
focuses on verbs (others focus on objects)
architecture is that of stories
data = characters
process = dramatic interactions
NPCs have lots of intelligence

no front end
overly complex process for building verbs
Mac only
poor documentation
time flow problems -- can’t get the right information to the right people
lacks critical mass of software
needs templates for interactions and verbs
no demo storyworld
built-in assumptions can be limiting
data formats are too closed
no viral sampling -- need to publicise more

10. Will Wright’s bizarre and brilliant ideas
players create story in sandboxes
open ended potential
successful in the marketplace
iconic content drives the drama (That is, the characters are so vague as to constitute icons of characters rather than true characters, leaving much to the player’s imagination.)

no authorial direction
hard to create drama
simplistic relationships and character models
missing many storytelling tools
needs a bigger sandbox

11. Higher-order plot approaches
emphasis on recombination

difficult to understand
shortage of details

TG: I did not fully understand the differences in this approach from the typical Erasmatron approach but generally agree with the philosophy of object reuse and recombination.

11a methods based on Georges Polti’s 31 Dramatic Situations
possibly useful for initialization
possibly useful in conjunction with character-driven methods

too high-level
doesn’t address the internals of stories

12. Story-driven games
already existing marketplace
well-understood methods

a well-defined marketplace that doesn’t really appreciate stories
The group was divided on whether story-driven games can evolve into true interactive storytelling. It is not architecturally set up to support drama.

TG: I believe this is the logical starting point. An iterative approach will be rewarded with the commercial success of Story-Game++ which allows the developer to continue with an open ended problem.

Overall Conclusions
Character-based and simulation methods are most likely to yield short-term results.

Purely plot-based methods are not in themselves adequate to support interactive storytelling; they would have more potential if we could translate them into computable terms.

Simulation-based platforms offer the strongest starting foundation for development, but the group has serious reservations about the potential of this strategy to yield good results for interactive storytelling. The biggest challenge is to shift focus toward dramatic considerations.

Dramatic sublanguages brilliantly solve the toughest problems of interactive storytelling, but do not begin to address plot issues.

The Erasmatron is the most-developed tool that directly addresses interactive storytelling. It is the ONLY tool that directly addresses interactive storytelling, It is impossible to fairly compare the Erasmatron with other strategies because they haven’t been built yet.

The Aarne-Thompson index of folktale motifs is a useful resource, but not an adequate design strategy.

Will Wright’s methods are not interactive storytelling; they stimulate and nurture the player’s storytelling imagination.

TG: I think the dream of Interactive Storytelling is still sometime off but I believe it will be achieved one baby step at a time. Darwin will prevail here and future history will record storytelling as part of the new breed of entertainment and this now impossible task with be a component in the entertainment software just like a graphic engine or physics engine There will be a story engine to keep the game interesting and 12 year olds around the world will toggle the checkbox off because itÕs so annoying when widow of the monster you just killed comes knocking on your castle crying because she has little ones to feed.

Chris Crawford’s own reactions to Phrontisterion are presented

The attendees of Phrontisterion IV, June 8-9, 2002

Standing, left to right: Piper Jackson in black top hat, Kevin Garlow in straw hat, Joseph Breitreiter in military officers hat, Kalev Tait in tall red and white hat, Todd Gemmell in tall blue hat, Jonathan Sari in red jacket, unrecognized person in helmet, Tim Emmerich in small hat, Riley Lynch in propeller hat, Benjamin Fallenstein in purple hat.
Kneeling, left to right: Laura Mixon in pink hat, her daughter Carita Gould, Chris Crawford in black hat (how appropo!) Rick Smith in brown hat, Emma Gould in green hat, Moose the Dog without any hat, Gordon Walton in engineer’s hat, Larry Mellon in green mad hatter’s hat.
Not shown: Huan Huya Chye