Blog Post

CHAT Report: Transforming Narratives

Soundbyte: Transforming Narratives
Wednesday, February 17, 2010
Hyde Hall, UNC-CH
CHAT Festival

This was a really excellent and illuminating talk!!

Into by John McGowan
Michael Young (Y)
Associate professor of computer science N.C. State University, director Liquid Narrative research group and co-director Center for Digital Entertainment. My relationship to narrative: focus is on computational models for interactive narrative. A principle mode of game interaction is through story. Virtual reality looks to recreate reality with high fidelity, but I look at the illusion of reality. Players understand this and enter into a compact with the system.  I want to diminish the role of the story designer, where the story is baked in, but rather the system and player come up with the story at run time. This has to appeal to a model for how narrative functions. Cognitive science, linguistic, film theory - try to build a computational model out of multidisciplinary analytic disciplines, and make it generative.

Computational building blocks. Narrative = story + discourse. Story is what happens inside the story world, including the actions that transpire over time. Discourse is not dialog, but the way the medium is used to convey the narrative (background music, graphics, etc). But, be careful, this view leads to reductionist engineering approach. Game narrative adds +interactivity. Building blocks are characters, setting, actions characters take. Intelligent design reasoning can be used as proxies for characters when storyboarding games.

A principal constraint: out of so much possibility, which actions are actually interesting (i.e. has tension, conflict, rising action)? Cognitive psychologists offer models of narrative comprehension. Game is tuned to trigger likely player cognitive reaction.

Discourse: content selection, organization, realization. Any narrative has elements that the author leaves out. Then how is it organized (camera movement, shot sequences, time shifting).
These are linguistic techniques to generate natural language stories. Camera planning is laid on top of story world to tell different aspects, i.e. manipulating players knowledge of the story.

Interactivity: coherence vs. control. Players feel like they have agency over the unfolding story. However, stories have a beginning, middle, and end. Unfettered interactivity works against that arc, and the complex dependencies youve built. Have to find balance, give the player the illusion of control while advancing the arc. Take all causal dependencies - which ones affect the grand narrative, which ones can break and which ones cant? Intervention - system steps in to prevent actions that would break the system. Else you have to throw away parts of your downstream plan and replace it with something else.

Cognitive models + AI models + narrative models > system building > evaluation (do the computational models actually embody the narrative thrust?)

Katherine Hayles (H)
Professor and director of graduate studies in the program in literature Duke University. Author of How We Think (2009).
Plot Aside: Narrative Strategies Beyond Causality: Probe or exploration on how to think about computational models that go beyond causality. That means plot is secondary. What kind of narrative strategies emerge that arent dependent on causality? Causal connections use a high cognitive mode.

Example from Michael Joyce, Twelve Blue. Uses proprietary Storyspace, using hypertext. Narrative as metonymic verbal images, which provide non-causal structure. The interface: eight bars, twelve strings. Narrative is played like an instrument. Graphic shows how threads are related, making links changes structures. Images are vivid, but we dont know how they relate. Following a hyperlink, we get another fragment. Eventually its possible to recuperate the plot, but we come to the inference of that plot from the images that form threads.

Example: Judd Morrissey, The Jews Daughter: Narrative as complex temporalities. A work in which we dont have much causality, but very complex temporalities because of the interface. Part of prose changes depending on what words you click. New part embedded in old part. Might make less sense. A verbal surface that is belated or premature. Temporality is not a single line but is always out of joint with its context. Daniel Dennett - multiple drafts model, difference between objective time and experienced time.

Kerry Lawrynovicz, Girls Day Out: Narrative as animated visual image. Letters rearrange themselves, enacted on the screen.

Y: 2 ways of categorizing H statements. Relation to interpretation process of player or viewer. Cognition involved beyond understanding the causal connections. Taking them out may cause dissonance, which could be OK, but there has to be a reason for it.

Audience questions:
Question: Do you have to have a finite number of possible interactions, or can you make computational narrative models that are completely open ended? For instance by trying to tell stories using people's personal media collections as story elements.
Y: Sequences that have properties, not just story elements. Building up inferences over time using metadata.

Question: Sequence: co-temporality? How do you handle in the meantime, something else was going on.
Y: Can be challenging to convey co-temporality. Representation and modeling/execution. Modeling is fairly easy because they allow for partial actions. Game cinematic can be more like film. Players can also experience two different storylines, revisionism, not viewed as pipeline.

Q: Causality, where does the scientific method come in?
Y: Great question, I form theory based on existing work and artifacts out there in the wild; then build model. Run humans through controlled experiment, run that back into the theory.
H: Tacit knowledge is still important even if it cant be articulated, especially for scientific process. State of initial confusion where youre bombarded with information you have to sort. The payoff is that when comprehension comes, it doesnt come as a linear understanding. You have to be willing to be confused and totally lost for a significant amount of time. When comprehension dawns, it is nuanced and multilayered, i.e. Gravitys Rainbow. Being bombarded. You have to just tolerate it! Can you model a situation where comprehension is delayed, or would that game be considered intolerable? Its the journey, not just the final realization.
Y: A lot of inferences are made about the intention of the storyteller. Amenable to systems that do this automatically. Looking at comprehension over time can be implicit or explicit: Im revealing this because I want you to retract that inference later.

Q: Thinking of culture shock, where you generate different things than you thought you were going to cause. Procedural and strategic thinking: the real way to win a game is to fail completely. How do you program to anticipate calculated disruptions?
Y: Puzzles to explain limitations. Manipulating expectations. Violating genre norms. Challenge is to back off from human encoding of that and place more responsibility on the system to play with that.

Q: Modeling drug interactions. Virtual models to learn something about real behavior in games?
Y: Yes, but not unintentionally. But its designed around understanding and options for action. You can manipulate whatever data youre working with, but from the systems perspective, its just stuff in the story. Not everythings a narrative, but thats a powerful way to understand stuff.

H: Good to see crossover between literature and computational models.
Y: Just the beginning in this field. Can be daunting to hear such articulate theories from other disciplines. Lots of potential.



Mike, I'm so glad you took notes because it took nearly all of my brain to contend with the fascinating questions being raised in this conversation.  For a non-scholarly person like me, it really illustrated the wonderful impact of the kind of conceptual exploration that can happen in an academic setting. It was a mental workout that left me intellectualy invigorated.

Kudos to the CHAT Festival on a great session!


This *was* a great session! And thanks also for the summary. I'm terrible at taking notes at talks, so this is a wonderful record.

I had some emerging inchoate thoughts during the talk, kind of rambling and weird, but here nonetheless:



Yeah, so far this has been my favorite session at CHAT. Narrative without causality and adaptive, automatic narrative generation are both game-changers. It was interesting to hear these two together. I think we've barely tapped the potential of these new forms of storytelling.