Narrative planning is the use of automated planning to construct, communicate, and understand stories, a form of information to which human cognition and enaction is pre-disposed. We review the narrative planning problem in a manner suitable as an introduction to the area, survey different plan-based methodologies and affordances for reasoning about narrative, and discuss open challenges relevant to the broader AI community.

This paper presents MULTISTYLE, a multi-agent centralized heuristic search planner that incorporates distinct agent playstyles to generate solution plans where characters express individual preferences while cooperating to reach a goal. We include algorithmic details, an example domain, and multiple different solution plans generated with unique agent playstyle sets. We discuss our intent to incorporate this planner in a tool for game level designers to help them anticipate and understand how teams of players with distinct playstyles may play through their levels. Ultimately, MULTISTYLE generates solution plans with a novel and increased expressive range by attempting to satisfy sets of action and proposition preferences for each agent.

A growing number of algorithms for story planning include the ability to create stories with failed actions -- in particular failed actions that occur because of the mistaken beliefs of the characters attempting them. To date, most of these systems have been evaluated analytically, primarily by comparing their expressive range to prior story generation systems. Empirical evaluation of these systems has been preliminary. In this paper, we outline a general comprehension-based approach to the evaluation of plan-based story generation. We describe how we specialize it for use evaluating story plans containing failed actions, and we describe the design and results of an experiment using this approach to evaluate plot lines produced by HeadSpace, a system that models the beliefs of characters and uses that model to generate plot lines containing actions that are attempted but that fail.

Previous work on story planning has lacked a knowledge representation for characters that attempt actions that fail because of the characters' misconceptions about the world state. This work describes HeadSpace, a state-space heuristic search planning system that generates stories that track and manipulate characters' beliefs about the story world. The planner produces story plans with actions that are attempted but fail. We show an example story plan that contains failed-action content that cannot be generated by typical planning-based approaches to story creation, and we provide an analytical evaluation that characterizes our planner's increased expressive range relative to other narrative planners addressing character belief and/or failed action execution.

Recent work extending planning algorithms that reason about action and change has been successful at supporting game design, player modeling, and story generation. Incorporating agent preferences over actions and propositions into a planning process allows for a more accurate prediction of what a human might do when solving a problem like playing through a game level. This paper presents the preference-based planning heuristic RPGPref which uses relaxed plan graphs (RPGs) and preference sets to guide a planner toward a preference-conforming path to its goal. A human subjects evaluation confirms that RPGPref successfully guides the planning process toward solution plans that recognizably match and differentiate player playstyles.

The General Mediation Engine is an architecture for producing game experiences with world mechanics drawn from a planning domain and problem. The system models possible player actions, state updates, and system responses in a graph representation derived from planner input. Paired with an interface that displays state and action information to the player, GME structures gameplay as an on-line expansion of its graph representation.

AIIDE 2014 is the second AIIDE event that has featured a playable experience track. This paper describes the seven entries that were accepted in the 2014 track, as well as the motivation behind the track and the criteria used to evaluate and accept entries.

Increasingly, research on narrative planning is expanding the expressive range of narrative generation systems, producing plot lines with structures like failed action, mistaken character belief and the integration of authorial and character-centered plans and intentions. Evaluation of these systems’ expressive capabilities is essential to determining their strengths. Some prior evaluative methods have measured the efficacy of narrative planners by characterizing a user’s experience during generated narratives. These approaches have focused on comparing the mental model a user forms during the experience of a narrative with the plan data structure that served as the basis of the narrative’s plot, but have not considered evaluating the user’s understanding of failed actions in narrative. To that end, we propose an algorithm to translate plans containing failed actions into a commonly used cognitive mode of narrative comprehension called QUEST. We then sketch how this translation will play a role in a planned evaluation of users’ experiences reading stories produced by narrative planning systems that generate stories with failed actions.

A goal of Experience Managers (EM) is to guide users through a space of narrative trajectories, or story branches, in an Interactive Narrative (IN). When a user performs an action that deviates from the intended trajectory, the EM uses a mediation strategy called accommodation to transition the user to a new desirable trajectory. However, generating the trajectory options then selecting the appropriate one is computationally expensive and at odds with the low-latency needs of an IN. We define three desirable properties (exemplar trajectories, narrative-theoretic comparison, and efficiency) that general solutions would possess and demonstrate how our plan-based Intention Dependency Graph addresses them.

In this paper we present the system called Bardic, which was developed over three years as the core technology in the Narrative for Sensemaking project, an effort to automatically generate narrative from low-level event data as an aid for sense making. At its core, Bardic is a narrative report generator that uses a logic-based language torepresent a story based on event/activity logs. Bardic generates different types of narrative discourse designed to convey different aspects of the underlying data. The system consists of aUnity application that allows users to explore generated stories and select parts of the stories to analyze in detail. In addition to its narrative generation capabilities, the application provides visualization and querycapabilities. In this paper we provide screen shots of the application in use, show examples of narrative output produced by Bardic, detail how Bardic generates its output, and discuss the current system's capabilities and limitations.

Interactive narratives (IN) are stories that branch and change based on the actions of a participant. A class of automated systems generate INs where all story branches conform to a set of constraints predefined by an author. Participants in these systems may create invalid branches by navigating the story world outside the bounds of the author's constraints. Two existing methods, choice removal and intervention, are designed to mitigate these situations. However, these methods are expected to lower invisibility, being recognized as system manipulations by the participant. In this paper we present an evaluation of a new method, domain revision, that is designed to have no negative effect on invisibility. We measure invisibility by asking survey participants how believable a choice's options and outcomes are in the context of a Choose Your Own Adventure story. We find that domain revision is more believable than choice removal when applied to a choice's options. We also find that domain revision is equally believable as intervention on a choice's outcomes because intervention does not cause a drop in invisibility.

Mediation is a plan-based interactive narrative generation algorithm that creates a cascading policy of plans. This policy represents a branching story and can be used by an execution manager in a game to control the series of events that unfold. With a few modifications we show that mediation's search space can embed all possible traversals through a game world. This shift allows gameplay to be modeled as an on-line mediation search where the game's interface is a representation of the underlying graph traversal. In this paper we outline these modifications and present a text-based implementation.

We present a metaphor through which to study games: games as conversation, which casts gameplay as a communicative exchange between player and game. We propose to view aspects of gameplay as speech acts, as defined by Austin and Searle, and we present several examples that illustrate the diverse locutionary, illocutionary, and perlocutionary acts present in the design of digital games. Through our perspective, we are able to cast problems relevant to the interactive entertainment community as discourse problems, where an interactive system must determine what to “say,” in order to elicit in the minds of players a specific mental model that will allow them to perform successfully in the game. We conclude with a research agenda that proposes to leverage the artificial intelligence paradigm of discourse planning to tackle the discourse problems of interactive entertainment.

A key part of managing a player's virtual reality experience is ensuring that the environment behaves consistently to the player's interaction. In some instances, however, it is important to change how the world behaves-i.e. the world's simulation rules or mechanics-because doing so preserves the virtual environment's intended quality. Mechanics changes must be done carefully; if too overt, they may be perceivable and potentially thwart a player's sense of presence or agency.This paper reports the result of a study, which demonstrates the widely-held but heretofore-untested belief that changing an environment's mechanics without considering what the player knows is visible to the player. The study's findings motivate the paper's second contribution: an automated method to perform invisible dynamic mechanics adjustment, which affords shifting a game's previously-established mechanics in a manner that is not perceivably inconsistent to players. This method depends on a knowledge-tracking strategy and two such strategies are presented: (1) a conservative one, relevant to a wide variety of virtual environments, and (2) a more nuanced one, relevant to environments that will be experienced via head-mounted virtual reality displays. The paper concludes with a variety of design-centered considerations for the use of this artificial intelligence system within virtual reality.

Storytelling presents a compelling context for the development of intelligent systems. Increasingly, research on intelligent systems has targeted the development of computational models for the generation and understanding of stories. However, few projects include in their accounts components that reflect insight in to the narrative comprehension process provided by narratology and cognitive psychology. In this paper, we synthesize these relevant perspectives into desiderata for computationally modeling the narrative sensemaking process. We describe the set of requirements that process models ought to satisfy should they aim to define a computational procedure reflecting the human sensemaking processes, either in the production of narrative or in its automated understanding. Storytelling is increasingly relied upon to entertain, educate, and engage American society in more compelling ways (Lenhart et al. 2008; Lohse et al. 2013; Duggan 2015). Driven by the increased demand to structure human interaction through stories and by the realization that artificial intelligence (AI) is well suited to doing so, the field of computational models of narrative (CMN) has progressed primarily on the two fronts that make up the narrative intelligence (Mateas and Sengers 1999) enterprise: computational narrative generation and computational narrative sensemaking (Mueller 2013). While distinct, these two areas are intrinsically related: the ultimate design criteria for a narrative artifact rests in the mind of the story consumer, and authors design the stories they tell to affect audiences in specific ways (Bordwell 1989; Sharples 1999). Thus, the field must advance these areas in tandem. Here, we focus on computational narrative sensemaking. Within the field of CMN, sensemaking1 has received the largest amount of attention in the literature to date; see the article by Mueller for a review of this work up until the year 2013. These systems have primarily tackled the challenges involved in understanding intuitively important narrative-theoretic concepts on the basis of sentential text; concepts include time, space, states, events, goals, plans, scripts, characters, and objects. Copyright c © 2019, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Historically, these narrative sensemaking systems have been referred to as story understanding systems. However, despite excellent work on developing computational models of narrative sensemaking, scholars – with notable exceptions – have not attempted to include in their accounts directly relevant perspectives from narratology and cognitive psychology. In this paper, we synthesize these relevant perspectives into desiderata for computationally modeling sensemaking. We describe the set of requirements that process models ought to satisfy should they aim to define a computational procedure that reflects or complements the human sensemaking processes conceptually described and empirically investigated by cognitive psychologists. We target modeling sensemaking in an online manner, i.e. during the consumption of the narrative. Importantly, we do not claim that AI for narrative sensemaking that reasons like humans do should be prioritized to the exclusion of other potential kinds of intelligent systems. While people remain our only examples of naturallyoccurring intelligent systems, there may exist many different psychologies that can be computationally characterized and may be relevant to different task environments (Doyle 1983). Operationalizing Narrative and Sensemaking By proposing to use computationally-precise definitions, we are implicitly making an ontological commitment over the structure of narrative and associated phenomena. To avoid leaving our ontological commitments tacit, we explicitly outline them here.

We introduce ShowRunner, a tool for visualizing story world execution within a 3D game environment. ShowRunner takes as input an abstract, declarative specification of a story script and a set of mappings between terms in the story and data elements in the game engine and executes the story’s actions, using virtual cameras to film and present the action to a user. The implementation details on the working of the tool, as well as instructions on how users with various design and API constraints can utilize the tool are discussed in this paper.

Deliberation-driven reflective sequences, or DDRSs, are cinematic idioms used by film makers to convey the motivations for characters adopting a particular course of action in a story. We report on an experiment where the cinematic generation system Ember was used to create a cinematic sequence with variants making different choices for DDRS use around a single decision point for a single character.

There have been a range of coding schemes to code story structure. However, few of these coding schemes map directly to expressive formal models of story that also characterize character beliefs or the complexities that arise when mistaken beliefs lead to action failure. We describe HeadCode, a coding scheme motivated by recent work in plan-based story generation.

An open challenge for AI in digital games is narrative experience management, the problem of automatically directing virtual characters in an interactive story to produce specific narrative effects for human participants. One important aspect of interactive narrative quality is participant choice, which provides the central distinction between interactive and linear storytelling. Most experience management work identifies storytelling patterns an agent should value or guides participants through high-value story trajectories. However, the types and quality of choices available in an interactive story are influenced not only by a model of narrative and specific story character actions, but also the world design that affords situations and actions. In this paper, we present a method for generating all unique, accessible choice combinations for a story world as a first step towards experience-driven interactive narrative world design. We benchmark the approach against several baselines and discuss its use as a tool for human and automated world designers.

Creation of machine generated cinematics currently requires a significant amount of human author time or manually coding domain operators such that they may be realized by a rendering system. We present FireBolt, an automated cinematic realization system based on a declarative knowledge representation that supports both human and machine authoring of cinematics with reduced authorship and engineering task loads.