Storylet-based narrative systems offer a flexible structure for interactive storytelling, but authoring a diverse set of meaningful storylets requires significant design effort. Large language models (LLMs) may have the potential to alleviate this burden by generating short-form narratives in response to predefined structural constraints for use in such systems. In this paper, we conduct a case study on ChatGPT 4o's ability to generate narrative fragments with parameterized intensity levels, and assess the alignment of user perceptions with these intended intensities. We hypothesize that newer large language models such as GPT-4o are able to effectively generate short-form narratives in response to parameterized inputs, and that user assessment of these LLMs' output will largely be in line with their intended intensities. To test this, we generate fifty stories across five unique story domains, each varying in a single parameter relevant to the story, such as success, damage to a relationship, chaos caused, or others, on a scale of 1 to 10. In a user study, participants were asked to compare pairs of stories with varying intensity differences (1,2, and 3) within each story domain. Analysis of user feedback, gathered through Likert-scale surveys, indicates a strong correlation between user-perceived intensity and the intended intensity ordering of generated stories. Additionally, user certainty in comparisons increased as the intensity gap between story pairs increased. The results support our hypothesis that LLMs can effectively generate stories with discernible intensity variations in response to parameterized input, and that users can reliably perceive these variations as intended. This research contributes to understanding the potential of LLMs in controlled narrative generation and provides support for their use in LLM-aided, hybrid narrative generation systems or games in the future.

This demo paper presents an interactive narrative game where players intervene in Shakespearean scenes to deescalate imminent violence using natural speech. The system employs a hybrid architecture: modern natural language processing (NLP) tools (speech-to-text, large language models, sentence transformers) interpret user utterances for known de-escalation strategies (e.g., active listening, redirection) and gauge their effectiveness, while symbolic systems model character emotional stances, beliefs, and personality traits that affect input interpretation, in order to select fitting responses. The project contributes an investigation of current NLP capabilities for interactive character-driven games and serves a pedagogical purpose, teaching conflict resolution and violence prevention within a broader educational framework.

Automatic story generation remains a challenging task, with modern storytelling systems continuing to lag behind human-authored works in quality—either constrained in plot planning or confined to narrow story domains. While contemporary narrative planners and storytelling systems demonstrate improvements, static story generation systems still lag behind. We present College Ruled, a mixed-initiative storytelling system iterating on past works in static story generation such as Lebowitz’s UNIVERSE and Skorupski’s Wide Ruled re-imagination of the author-goal-based model of story generation. An effort to fold in the best qualities of modern storytelling systems into a static story generation context, College Ruled utilizes story waypoints, a drama management system, and causality weighting to guide plot fragment selection. This allows for authorial influence over how character relationships are shaped and provides agency over the dramatic arc of a spun tale. In conjunction with a stochastic modification to the A* algorithm, College Ruled offers a framework to be used alongside a library of manually authored plot fragments to produce rich variety in generated tales while meeting narrative goals. We describe the implementation of our system and then investigate its effectiveness in drama management and producing stories that meet author specifications.

Common approaches to behavioral artificial intelligence, like behavior trees, utility-based approaches, and machine learning are often lacking with regards to expressivity of the decision-making process. Behavior trees are too rigid; utility-based approaches are focused on outcomes, not process; and machine learned agents seek only to maximize reward and minimize loss. A combination of AI approaches, however, may provide more nuanced, visible, and coherent processes that can be realized through the performance of character behaviors. In this extended abstract, we present Viv: a system that blends defeasible logic programming with dynamic behavior trees. To our knowledge, no extant works have bridged the gap between defeasible reasoning and action, particularly in regards to the performances of virtual nonplayer characters. We begin by explaining the design of our system, which was heavily influenced by the Stanislavski method of acting. We then describe the technical implementation of our system. Finally, we describe future directions of research that could provide benefit to, or benefit from, our system.

Many genres of video games value the fidelity of background non-player characters (NPCs) for they create the illusion of life necessary to support a believable world. However, in some genres of games, there is often a layer of abstraction between player actions and how NPCs process and respond to those actions. In this work, we argue that a system of computational gossip, grounded in sociological and social-psychological theory, can reduce that abstraction and improve believability of NPCs by deepening an NPCs reasoning of player action. In the form of a case study, this work presents the game Sunset Valley: an early step towards a system of computational gossip. Then by inspecting Sunset Valley’s gossip model, we analyze our game by exploring several well-established sociological and social-psychological frameworks, identify which frameworks are well-represented in Sunset Valley, and which frameworks are not. Finally, we conclude to what extant artificial intelligence systems can be used to address the shortcomings of our system.

Game mechanics, the affordances they create for players, and the agency experienced by players as a result of interacting with those mechanics are all, indisputably, intertwined. This work seeks to explore the relationship between game mechanics and agency by way of building an agency-centered ontology of video game mechanics. Building on existing framings of agency, we devise a set of agency-centered domains, properties belonging to those domains, and questions about game mechanics that reveal those properties. We build a dataset of game mechanics from different games and, using our domain-specific set of questions, illuminate properties of those game mechanics. Finally, we explore how those properties are interrelated, and identify examples of game mechanics that, according to our analysis, inspire the most agency.

In this work we demonstrate that the First Order Model Agnostic Meta Learning (FOMAML) algorithm trained on the Hanabi Open Agent Dataset (HOAD) results in a model that is able to outplay both a naive MLP baseline, as well as a randomly selected partner in the Hanabi Ad-Hoc Challenge, in both low-shot and zero-shot setups. We first show that HOAD is well suited for the meta-learning task because its agents are high quality and utilize diverse strategies, thereby confirming that MAML is generalizing, and not memorizing agent strategies. We then detail our application of FOMAML to the cooperative decision making problem Hanabi entails, and we also provide evidence supporting recent results that the task update of MAML gives little to no test time performance boost. The pretrained models and game data are made available online at https://github.com/aronsar/hoad.

Increasing the behavioral nuance and interactivity of non-player characters in story worlds comes with a growing cost in the time and effort expended by authors. This paper proposes a system, Cast Affinity Satisfiability Toolkit (CAST), which uses answer set programming to lessen this burden while supporting author autonomy. CAST uses user-defined constraints about personality facets and interests to generate a cast of characters and their relationships. Once salient personality facets and character interests are defined, additional constraints may be supplied about character similarity and affinity. This refines not only character personality but the social connections between characters. Rules about affinity, or the attraction characters feel towards each other, may be used to weight personality facets and interests to generate consistent characters and their relationships that consider the significance of these traits. Although the quality of output is contingent upon the quality of constraints, CAST respects all author supplied constraints and provides the framework to generate NPCs that are consistent and whose relationships are tailored to reflect what is significant in the story world.

Wave Function Collapse (WFC) is a non-backtracking, greedy search algorithm that is commonly known for its ability to take an example image and generate similar images. Since its inception, technical artists have explored the algorithm's extensibility and usability through various implementations spanning from 3D world generation to poetry creation. However, there has been no integration of design constraints into the generative process. In this paper, we explore WFC as a constraint satisfaction solver to integrate design principles and practices by modifying components within the algorithm. First, we extend the local constraint reasoning by incorporating non-local constraints as well as upper and lower bounds. Next, we further manipulate the generative space by introducing weight recalculation and dependencies. Lastly, we evaluate our design-focused variant of WFC against the original implementation to examine the associated costs in computational time and memory usage. In summary, this paper describes a technical implementation of integrating design constraints into WFC and analyzes the computational trade-offs.

Asset generation has a huge time cost associated with it in games. In order to reduce this time cost, designers have adopted Procedural Content Generation (PCG) systems into their workflows. However, most PCG techniques may only give a modest workflow speed up. While some PCG techniques can give immense speed ups, they have extra time costs that might not be worth it. These techniques require users to encode their domain knowledge in a way these techniques can understand. This can be a costly process and the technique might not be able to be reused in another project. We propose a framework for encoding domain knowledge in a way to promote reusability. We provide an example that is based in architectural design philosophies.

While game narrative provides a story for the player to experience, the moment-to-moment decisions made by the player are just as important to the experience. These decisions make up a personal narrative that the player creates through their choices and actions within the game. These stories describe the player's experience, and are the stories that often get shared and retold by the player. However, these narratives are rarely captured by the game and instead rely on the player to memorize and retell them. In response to this, we designed Loominary, a game platform that plays Twine games using a rigid heddle table-top loom as a controller. Not only does this provide a new method of interacting with a game, but also records each of the player's choices into a tangible object that is created by interacting with the game. Loominary is a working prototype and in this paper, we discuss the design considerations and areas for improvement.

Abstract—This paper presents Comme il Faut (CiF), an artifi-cial intelligence system thatmatches character performances to ap-propriate social context, with the goal of enabling authors to write high-level rules governing expected character behavior in given so-cial situations, rather than specific fixed choice points in a curated narrative structure. CiF models characters with a complex set of traits, feelings, and relationships, who can form intents, take ac-tions, relate to a shared cultural space, and remember and refer to past events. A set of authored rules encoding appropriate be-havior within a specific story world allow these characters to se-lect actions to take (and respond to actions by others) in a manner consistent with their own personal and social concerns as well as a shifting interpersonal context. Through the development and re-lease of PromWeek, a complete game using CiF as its narrative en-gine, we show how the system successfully creates complex narra-tives that are unique for each player and directed by those players’ attempts to make progress towards story goals. We also show how CiF continues to be used in several in-progress interactive experi-ences (Mismanor and IMMERSE), speaking to the utility and flex-ibility of its design. Index Terms—Artificial intelligence, emergent narrative, game design, interactive drama. I.

Prom Week places players in a typical high-school, abuzz with excitement over the upcoming prom. Players indirectly sculpt the social landscape by having these hapless highschoolers engage in social exchanges with each other. The results of these social exchanges are many and varied---ranging from mild fluctuations in respect to characters professing their eternal love for one another---and are informed by over 5,000 sociocultural considerations encoded in first order logic. Through massaging the interpersonal relationships and learning the personal intricacies of the characters, the player can solve a series of social puzzles; such as making the class-nerd the Prom King, or bringing peace between feuding jocks and preppies.

In this paper, we present Prom Week, a social simulation game about the interpersonal lives of a group of high school students in the week leading up to their prom. By starting the design of the game with a theory of social interaction, Prom Week is able to present satisfying stories that reflect the player's choices in a wide possibility space -- two features that rarely accompany one another. This paper reports the design details of how Prom Week utilizes social physics to achieve rich character specificity while maintaining a highly dynamic story space.