Tuesday, 22 December 2015

ICIDS and Interactive Fiction

ICIDS The International Conference on Interactive Digital Storytelling (ICIDS) has been held annually since 2008, or 2001 if you count the two conferences it was formed from (ICVS and TIDSE).

The papers from each conference have been collected into a set of PDF (and physical) books. These are quite expensive to download, but working at a university means I have free access to them at the moment.

Topics include things unrelated to IF such as animation and virtual interfaces, but there are many subjects related to IF: narrative theory, story generation, agency, player emotions, drama management, autonomous characters, emergent narrative and natural language.

I haven’t been to an ICIDS conference, but you can read a first hand account from Juhana Leinonen who attended in 2015 and Emily Short who was invited to talk about Versu in 2014.

Here are the papers I’ve come across that are relevant to this blog, along with links where I could find them.

The Conferences

ICIDS was formed from two pre-existing conferences, the International Conference on Virtual Storytelling (ICVS) and Technologies for Interactive Digitial Storytelling and Entertainment (TIDSE).

Only a subset of the papers is listed below. If you are interested the in other papers too, click on one of the Springer LNCS links on the ICIDS home page for a conference’s table of contents, then search for a PDF version of a paper – most of them exist somewhere on the web. There is no Springer link for TIDSE 2003, but the contents is listed here.


Narratology for Interactive Storytelling: A Critical Introduction (pdf)
(Marc Cavazza and David Piz, TIDSE 2006)

Overview of narrative theories that have been used in Interactive Drama research.

Interactive Narrative, Plot Types, and Interpersonal Relations (pdf)
(Marie-Laure Ryan, ICIDS 2008)

Basic plot types (epic, epistemic and dramatic) and how they can be applied in an interactive setting; aesthetic goals of an interactive narrative; emotional immersion.

Bringing Interactivity into Campbell’s Hero’s Journey (pdf)
(Guylain Delmas, Ronan Champagnat, and Michel Auger, ICVS 2007)

Our work relies on the use of emergent narrative theory, with a plot monitoring architecture that observes the course of the game, analyses players’ actions and gives directives to the game in order to produce narrative. We make decisions based on a global vision of narrative instead of a local one.

Story, Plot and Character Action: Narrative Experience as an Emotional Braid (link)
(S. Nath, TIDSE 2003)

Developing a model for emergent narrative. The beginning is a bit wordy, but section 4 on the narrative experience has some thoughts about theme, emotion, mood, plot and character.

Metrics for Character Believability in Interactive Narrative (pdf)
(Paulo Gomes, Ana Paiva, Carlos Martinho and Arnav Jhala, ICIDS 2013)

A study on what makes a character believable: behavior coherence, changing with experience, awareness, behaviour understandability, personality, emotional expressiveness, social relationships with other characters, visual impact, predictability.

Emergent Narrative

Purposeful Authoring for Emergent Narrative (pdf)
(Sandy Louchart, Ivo Swartjes, Michael Kriegel and Ruth Aylett, ICIDS 2008)

An interactor within an Emergent Narrative determines the direction of the narrative development by engaging in certain interactions, and in turn the narrative development constrains the probable future interactons.

In EN we try to remove the need to think in terms of plot. To this end, an EN system models how characters come to make certain dramatic decisions, so that the author is then left to determine the content of the emergent narrative, which is raised to a more declarative level. This way, the author can think directly in terms of these interactions and what happens locally, rather than in terms of the plots that should occur.

We suggest that authoring for EN is a continuing process involving finding dead ends and resolving them by authoring new content for that situation.

Let’s Pretend I Had a Sword: Late Commitment in Emergent Narrative (pdf)
(Ivo Swartjes, Edze Kruizinga, and Mariët Theune, ICIDS 2008)

The idea of late commitment is based on the observation that in improvisational theatre, actors do not work with a predetermined, agreed upon story world, but frame it as they go along. We currently use late commitment in two cognitive processes. First, goal management has been extended to allow the character agents to justify adoption of new goals when they have no goals to pursue.

Second, agents may create plans for their goals. If the captain has adopted a goal to find out whether the approaching ship is friend or foe, he can make a plan involving looking through binoculars. If the story world does not contain binoculars yet, they can be framed to be in the captain’s cabin, affording a plan for the captain to go to his cabin to get them.

Open Design Challenges for Interactive Emergent Narrative (pdf)
(James Owen Ryan, Michael Mateas and Noah Wardrip-Fruin, ICIDS 2015)

The problems of modular content (independent and recombinable content), compositional representational strategies (ability to express internal processes), story recognition and story support.

What Makes a Successful Emergent Narrative: The Case of Crusader Kings II (google books link)
(Bertrand Lucat and Mads Haahr, ICIDS 2015)

Every autonomous agent in the game, as well as the player’s current character, are characterized by a number of traits which carry opinion modifiers that affect a scale of -100 to 100 in characters’ mutual opinions of one another. Complementary and opposing traits create tensions between the various agents and the player’s character, which in turn drives a great deal of intrigue, betrayal, and conflict.

The depth, yet ultimate simplicity and transparency, of the trait system is essential to the player’s engagement with the game’s emergent narrative.

Plot Fragments

See also this talk by Ken Levine at the 2014 GDC conference on narrative legos and these articles on story bricks.

Story Planning with Vignettes: Toward Overcoming the Content Production Bottleneck (pdf)
(Mark O. Riedl and Neha Sugandh, ICIDS 2008)

We use the term vignette to refer to a fragment of a story that represents a “good” example of a situation and/or context that commonly occurs in stories. For example, a library of vignettes would contain one or more specific instances of bank robberies, betrayals, cons, combat situations, etc.

Vignettes are fragments of story structure. We use the minimal vignette rubric: a minimal vignette is one in which removing any one action from the vignette causes it to no longer be considered a good example of the situation and/or context it was meant to represent.

Schemas in Directed Emergent Drama (pdf)
(Maria Arinbjarnar and Daniel Kudenko, ICIDS 2008)

The director overlooks the emergence of the drama and uses schemas to direct the drama by giving the actors appropriate schemas to play out. The schemas do not occur in a strict sequential order. Instead the schemas are overlapping and actors typically play out multiple schemas at any given time.

The drama can not move between acts until the objectives of the acts have been adequately satisfied. As an example, the drama will not move from Act I to Act II until characters’ key characteristics have been exposed. If a character is to be intelligent, playful and curious then she needs to have played out actions that are intelligent for a value above a given threshold T, and the same for playfulness and curiosity.


Planning Formalisms and Authoring in Interactive Storytelling (pdf)
(Fred Charles, Miguel Lozano, Steven J. Mead, Alicia Fornes Bisquerra and Marc Cavazza, TIDSE 2003)

AI planning formalisms have been one of the main techniques supporting interactive storytelling, as planning precisely deals with the generation of action sequences fulfilling a goal. Key elements of the storyline appear as the main goals of a feature characters.

Traditional models in narratology describe a course of action quite incompatible with the smooth unfolding of the main character’s plan. This means that plan failure is an essential aspect of narrative interest. While the overall goal can be eventually achieved, dramatic interest can be obtained by “local” plan failure and subsequent re-planning or plan repair.

Believable Agents and Intelligent Story Adaptation for Interactive Storytelling (pdf)
(Mark O. Riedl and Andrew Stern, TIDSE 2006)

We represent narratives as partially-ordered plans. A plan contains steps – events that change the state of the world – and annotations that explicitly mark the temporal and causal relationships between all steps in the plan, defining a partial ordering.

Using planning structures to model scenarios is advantageous because a plan can be analyzed for points in which failure can occur. We analyze the causal structure of the scenario to determine all possible inconsistencies between plan and simulation state that can occur during the entire duration of the scenario. For every possible inconsistency that can arise that threatens a causal link in the plan, an alternative scenario plan is generated.

Controlling Narrative Generation with Planning Trajectories: The Role of Constraints (pdf)
(Julie Porteous and Marc Cavazza, ICIDS 2009)

Characteristics of a “good” plan, such as optimality, aren’t necessarily the same as those of a “good” narrative, where errors and convoluted sequences may offer more reader interest, so some narrative structuring is required.

We can introduce conflict into the narrative by imposing constraints that draw the narrative in directions that set the goal of one of the central characters against the others.

Drama Management

Failing Believably: Toward Drama Management with Autonomous Actors in Interactive Narratives (pdf)
(Mark O. Riedl and Andrew Stern, TIDSE 2006)

The drama manager adapts its narrative automatically to the actions of the user. The trade-off for using this approach is that narrative content cannot be known a priori, so special attention must be paid to the coordination between the drama manager and believable agents.

Sometimes a believable agent is required to fail to achieve a goal, in the interest of preserving the drama manager’s plotline. We use a technique where a believable agent can (a) intelligently and believably avoid situations where the agent and drama manager come into conflict, and (b) enact a plausible and believable explanation for goal failure when conflict with the drama manager is unavoidable.

Developing a Drama Management Architecture for Interactive Fiction Games (pdf)
(Santiago Ontañón, Abhishek Jain, Manish Mehta and Ashwin Ram, ICIDS 2008)

Adding a Drama Manager to the game Anchorhead by Michael S. Gentry.

The game author specifies a set of drama manager actions. These actions represent the things that the drama manager can carry out to influence the game, e.g. “prevent the player from entering the library by locking the door” or “make the bar keeper start a conversation with the player about the suspicious person.” DM actions can be classified into several groups: hints, causers (force something to happen), deniers (prevent an action), temporary deniers and reenablers (reenable a previously denied line of action).

In order to decide which DM actions to execute, the Drama Mangement Module performs a search in the space of possible stories. Such search is represented as a search tree with alternating DM actions and player actions. The DMM will select the DM action that maximizes interestingness.

Directorial Control in a Decision-Theoretic Framework for Interactive Narrative (pdf)
(Mei Si, Stacy C. Marsella and David V. Pynadath, ICIDS 2009)

Thespian agents possess a “Theory of Mind” and can model emotions and social normative behaviors. The Theory of Mind capacity allows the agents to reason about others’ beliefs, goals and policies when deciding their own actions.

A director agent is designed to monitor the progress of the story, predict its future development and adjust the virtual characters’ behaviors and beliefs if necessary to prevent violations to directorial goals. Such goals are used by the author to indicate how they want the story to progress, such as when an action should happen or a character should change its beliefs about another character.

A Simple Intensity-Based Drama Manager
(Christopher Ramsley, Matthew Fugere, Randi Pawson, Charles Rich and Dean O’Donnell, ICIDS 2010)

PDF unavailable, but there is information about the Drama Manager component of the Wind’s End game here (2.4M PDF; see pages 38-41).

Typical difficulties with a totally autonomous goal-driven architecture include either too few or too many interesting things happening at the same time, and the interesting events not happening where the player is currently located.

The basic idea of the Drama Manager is to enforce a rising arc of dramatic intensity. Each character goal is manually assigned an intensity from 0 (unimportant) to 5 (climactic). The Drama Manager allows at most one character to have the currently most intense goal at any moment in time, and incrementally ramps up the intensity of that highest goal.

Particular Aspects of Narrative

Adaptive Narrative: How Autonomous Agents, Hollywood, and Multiprocessing Operating Systems Can Live Happily Ever After (pdf)
(Jay Douglas and Jonathan Gratch, ICVS 2001)

Foreshadowing in an interactive environment, and having background narratives that can be brought into the foreground.

A Use of Flashback and Foreshadowing for Surprise Arousal in Narrative Using a Plan-Based Approach (pdf)
(Byung-Chull Bae and R. Michael Young, ICIDS 2008)

Prevoyant aims to create surprise at an important story outcome, which can make the reader more engaged in the story. For the surprise arousal, Prevoyant uses two narrative techniques: foreshadowing and flashback. Foreshadowing provides the reader with anticipation; flashback explains what caused the surprise (unexpected outcome) to happen.

Narrative Generation for Suspense: Modeling and Evaluation (pdf)
(Yun-Gyung Cheong and R. Michael Young, ICIDS 2008)

Suspense is the feeling of excitement or anxiety that audience members feel when they are waiting for something to happen and are uncertain about a significant outcome. A reader’s suspense level is affected by the number of solutions available to the problems faced by a narrative’s protagonists. The reader’s suspense is heightened when undesirable outcomes are likely to happen over preferred outcomes.

Suspenser takes three elements as input: a fabula (story world), a given point t in the story plan that corresponds to the point where the reader’s suspense is measured, and the story length desired by the system user. The system then determines the sjuzhet, the content of the discourse to be used to convey the story up to t to a reader, which enables the reader to infer a minimum number of complete plans for the protagonists’ goal, following the psychological research on suspense.

Backstory Authoring for Affective Agents (google books link)
(Stefan Rank and Paolo Petta, ICIDS 2012)

Backstories are events that happened before or outside the main plot that impact how and why a character behaves in a certain way. We present an outline of how this idea can be realised for one particular affective agent architecture, ActAffAc, enabling the configuration of characters by a selection of short action sequences as backstory events.

Personality, Emotions and Autonomy

An Emotional Architecture for Virtual Characters (link)
(Ricardo Imbert and Angélica de Antonio, ICVS 2005)

Cognitiva, a generic architecture for characters with emotional behavior.

Personality Templates and Social Hierarchies Using Stereotypes (pdf)
(Robert Mosher and Brian Magerko, TIDSE 2006)

Defining characters using stereotypes; the Five Factor Model of personality; power relations within social hierarchies.

Revisiting Character-Based Affective Storytelling under a Narrative BDI Framework (pdf)
(Federico Peinado, Marc Cavazza and David Pizzi, ICIDS 2008)

Using the Belief-Desire-Intention psychological framework to model characters; extensions to the BDI model in a narrative context.

Emotional Appraisal of Moral Dilemma in Characters (google books link)
(Cristina Battaglino and Rossana Damiano, ICIDS 2012)

Moral emotions, like pride and shame, arise as a consequence of moral values; their intensity depends on the importance of the values they are related with and is increased if these values are shared with other agents.

The aim is to create artificial characters who display the emotional range triggered by the occurrence of a moral dilemma, defined as mutually exclusive options that put different values at stake. Remorse is also part of the game, when the character, after deliberating and acting, is left alone with her conscience.

Freedom of Action

Player-Protagonist Motivation in First-Person Interactive Drama (pdf)
(Jeff Rawlings and Joe Andrieu, TIDSE 2003)

The Personal Narrative Agent (PNA) observes the player in the simulation and presents opportunities to begin stories based on the player’s unconstrained interactions. Players’ choices act as a self-selection mechanism ensuring that players experience stories of interest given their individual motivations and goals.

Players may fail to notice a story opportunity, or fail to understand how to act on it. Alternately, the narrative system may incorrectly interpret the player’s actions. To counter both challenges, the PNA confirms the player’s intent by presenting multiple escalating hypotheses to the player. This escalation reiterates the consequences of the player’s chosen action, verifies his commitment to those consequences, and advances the story accordingly.

Adaptive Storytelling and Story Repair in a Dynamic Environment (google books link)
(Richard Paul, Darryl Charles, Michael McNeill and David McSherry, ICIDS 2011)

An important challenge is the need for stories to be seamlessly adapted when story plans are invalidated by unforeseen events in the game world, such as the actions of Player Characters.

We consider two alternative approaches to detecting invalid plan steps while a story is in progress. The first is simply to look one step ahead to check that the preconditions of the next plan step are satisfied. The second is to continuously check the preconditions of all future plan steps. This increases the chance of finding a consistent plan repair because it enables the story manager to avoid commitments being made by story characters close to the point of failure.

The repair method is to search for an alternative story plan that begins with the same steps as the original plan, up to and including the most recent step that has already been completed.

Natural Language

Natural Language Understanding in Façade: Surface-Text Processing (pdf)
(Michael Mateas and Andrew Stern, TIDSE 2004)

Processing player input in the Façade game.

What Does Your Actor Remember? Towards Characters with a Full Episodic Memory (pdf)
(Cyril Brom, Klára Pesková and Jiri Lukavsky, ICVS 2007)

Characters being able to summarise past events; modelling character memory; forgetting less emotionally important episodes.

Combinatorial Dialogue Authoring (pdf)
(James Owen Ryan, Casey Barackman, Nicholas Kontje, Taylor Owen-Milner, Marilyn A. Walker, Michael Mateas, and Noah Wardrip-Fruin, ICIDS 2014)

We present an annotation scheme and combinatorial authoring procedure by which a small base of annotated human-authored dialogue exchanges can be exploited to automatically generate many new exchanges. The combinatorial procedure builds recombinant exchanges by reasoning about individual lines of dialogue in terms of their mark-up, which is attributed during annotation and captures what a line expresses about the story world and what it specifies about lines that may precede or succeed it in new contexts.


For an overview of this topic, see Review of Ontology Based Storytelling Devices (pdf).

Formal Encoding of Drama Ontology (pdf)
(Rossana Damiano, Vincenzo Lombardo, and Antonio Pizzo, ICVS 2005)

“Drammar” theory of drama, including an analysis of Hitchcock’s North by Northwest.

» Representing Dramatic Features of Stories through an Ontological Model (link)
(Mario Cataldi, Rossana Damiano, Vincenzo Lombardo, and Antonio Pizzo, ICIDS 2011)

More on the Drammar theory of drama (in the context of story understanding, not generation).


Authoring Highly Generative Interactive Drama (pdf)
(Nicolas Szilas, Olivier Marty and Jean-Hugues Réty, ICVS 2003)

The IDtension system.

The field of Interactive Drama has been historically related to the field of believable agents, the simulation of characters evolving in a virtual world. However, simulating human beings with intelligent agents is not sufficient to generate a well constructed narrative, because fundamentally, narrative contains “inverse causality.”

» IDtension: A Narrative Engine for Interactive Drama (pdf)
(Nicolas Szilas, TIDSE 2003)

More on the IDtension system, with some thoughts on the laws of narrative and modelling of obstacles.

Integrating Plot, Character and Natural Language Processing in the Interactive Drama Façade (pdf)
(Michael Mateas and Andrew Stern, TIDSE 2003)

The Façade system. There is a more in depth paper here.

An architecture that integrates emotional, interactive character behavior, drama-managed plot and shallow natural language processing. Authors can organize hierarchies of reactive behaviors and natural-language discourse contexts into entities called story beats; a drama manager globally sequences beats chosen from a large pool in order to make story tension rise and fall.

Character-Focused Narrative Generation for Execution in Virtual Worlds (early draft pdf)
(Mark O. Riedl and R. Michael Young, ICVS 2003)

In general, narrative generation systems that generate highly coherent narrative structures often neglect issues of character and believability. Likewise, systems that capitalize on the use of highly believable characters tend to promote poor narrative structure.

The Actor Conference (ACONF) system is explicitly designed to take advantage of the strengths of both the character-centric and author-centric techniques and thus achieve both strong plot coherence and strong character believability.

The Virtual Storyteller: Story Creation by Intelligent Agents (pdf)
(Mariët Theune, Sander Faas, Anton Nijholt, and Dirk Heylen, TIDSE 2003)

A non-interactive story generation system.

In the Virtual Storyteller, plots are not pre-defined but created by the actions of the characters, guided by a virtual director. For a clear task division within the system, we choose to use a separate director agent who has general knowledge about plot structure, rather than giving such knowledge directly to the characters.

Both characters (or “actors”) and director are implemented as intelligent agents, capable of reasoning within their own domain of knowledge. The characters can make plans to achieve their personal goals using story-world knowledge. The director is able to judge whether a character’s intended action ts into the plot structure, using both story world knowledge and general knowledge about what makes a “good” plot.

» A Fabula Model for Emergent Narrative (pdf)
(Ivo Swartjes and Mariët Theune, TIDSE 2006)

More about Virtual Storyteller. Generating plot from fabula (causally connected elements of a story).

No comments:

Post a Comment