Elsewise : Authoring AI-Based Interactive Narrative with Possibility Space Visualization CCS: Computing methodologies Natural language processing CCS: Applied computing Computer games CCS: Software and its engineering Interactive games
Yi Wang, John Joon Young Chung, Melissa Roemmele, Yuqian Sun, Tiffany Wang, Shm Garanganao Almeda, Brett A. Halperin, Yuwen Lu, Max Kreminski · 2026
Read Elsewise : Authoring AI-Based Interactive Narrative with Possibility Space Visualization CCS: Computing methodologies Natural language processing CCS: Applied computing Computer games CCS: Software and its engineering Interactive games by Yi Wang, John Joon Young Chung, Melissa Roemmele, Yuqian Sun, Tiffany Wang, Shm Garanganao Almeda, Brett A. Halperin, Yuwen Lu, Max Kreminski (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-15295.txt.
Introduction
Yi Wang, John Joon Young Chung, and co-authors build a tool for writing interactive stories with AI. Their 2026 arXiv paper deals with interactive narrative. Authors craft a space of possible story paths, and the player's input picks which path becomes real.
Generative AI can improvise fresh content inside that space as the player acts. The authors present Elsewise, an authoring tool for this kind of work. It shows the possibility space as a visual map. Writers can see the branches their story offers and where the AI may take them.
Catalog record
| CREATOR | Yi Wang, John Joon Young Chung, Melissa Roemmele, Yuqian Sun, Tiffany Wang, Shm Garanganao Almeda, Brett A. Halperin, Yuwen Lu, Max Kreminski |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 103 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-15295
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Elsewise : Authoring AI-Based Interactive Narrative with Possibility Space Visualization CCS: Computing methodologies Natural language processing CCS: Applied computing Computer games CCS: Software and its engineering Interactive games Yi Wang, John Joon Young Chung, Melissa Roemmele, Yuqian Sun, Tiffany Wang, Shm Garanganao Almeda, Brett A. Halperin, Yuwen Lu, Max Kreminski Abstract . Interactive narrative (IN) authors craft spaces of divergent narrative possibilities for players to explore, with the player’s input determining which narrative possibilities they actually experience. Generative AI can enable new forms of IN by improvisationally expanding on pre-authored content in response to open-ended player input. However, this extrapolation risks widening the gap between author-envisioned and player-experienced stories, potentially limiting the strength of plot progression and the communication of the author’s narrative intent. To bridge the gap, we introduce Elsewise : an authoring tool for AI-based INs that implements a novel Bundled Storyline concept to enhance author’s perception and understanding of the narrative possibility space, allowing authors to explore similarities and differences between possible playthroughs of their IN in terms of open-ended, user-configurable narrative dimensions. A user study (n=12) shows that our approach improves author anticipation of player-experienced narrative, leading to more effective control and exploration of the narrative possibility spaces. Figure 1. We propose the novel concept of “Bundled Storylines” for perception and sensemaking of narrative possibility space in AI-based interactive narrative authoring. Given a dataset of player-experienced storylines and an author-configurable collection of narrative dimensions, we “bundle together” narrative states that are indistinguishable in terms of the given narrative dimensions to construct branching storyline structures. We implemented Bundled Storyline Visualization (BSV) in an AI-based IN authoring system called Elsewise . Authors can customize the visualization to their needs by (a) creating BSVs from data-derived dimensions, (b) creating BSVs from user-defined dimensions, and (c) creating BSVs that cross over dimensions from existing BSVs. Leveraging BSV, authors can intuitively shape the distribution and trajectories of narrative states for their storytelling needs. This figure introduces a Bundled Storyline Visualization (BSV) system designed for understanding and navigating narrative possibility spaces in AI-based interactive storytelling. The figure demonstrates how narrative states that share similar characteristics can be bundled together to create comprehensible branching structures from complex story data. The top portion illustrates the conceptual workflow. Starting from experienced storylines shown as tangled gray paths and the narrative intents ("A story where courage leads to happiness"), the system extracts narrative dimensions such as emotion level and location (indicated as clips). Protagonist happiness is a dimension that is reused over again within this figure. The author themselves can also define their own dimension (protagonist courage, b). These dimensions are then used to create a bundled branching storyline representation on the right, where similar narrative paths are grouped by color-coded dimensions (also represented as clips), creating a more organized tree-like structure with dimension labels. The bottom section, c, shows the authoring process across three stages. Using protagonist happiness and protagonist courage as core dimensions, each stage shows a two-dimensional space plotting "not courageous" to "courageous" on one axis and "not happy" to "happy" on the other, with colored dots representing different narrative states. The progression shows how authoring refines the narrative space: starting with much concentrated states, to the versions where the story is progressing from not happy-not courageous to happy-courageous. 1. Introduction Interactive narrative (IN) experiences permit players to explore a space of divergent narrative possibilities crafted by an author, with the player’s input determining which narrative possibilities they will actually encounter ( Riedl and Bulitko 2013 ; Green and Jenkins 2014 ) . The most prominent form of IN features multiple branching storylines reflecting the consequences of different player actions. A fundamental challenge for IN authoring lies in the tension between two competing needs: the need for coherent expression of the author’s narrative intent ( authorial intent ) and the imperative to immerse audiences in the storyworld by granting them meaningful agency over plot progression ( player agency ) ( Riedl and Bulitko 2013 ) . Conventional IN authoring workflows address this challenge by predefining the range of possible player actions and creating storylines tailored to each potential course of action. However, comprehensively authoring appropriate storylines for many different potential sequences of player actions can be prohibitively burdensome, limiting the degree of player choice typically supported by conventional IN systems. Many authoring tools ( Green et al. 2021 ) have been introduced to help IN authors overcome this limitation, but the authorial burden ( Jones and Millard 2024 ) associated with high-agency forms of IN remains substantial. Large language models (LLMs) show potential to transform IN by enabling free-form player input and generating dynamic narrative content tailored to player actions in real-time. Traditional interactive narratives rely exclusively on pre-authored storylines created during development, limiting players to predetermined narrative paths. In contrast, LLM-powered systems can synthesize the author’s narrative intent with player input during gameplay, dynamically generating personalized narrative experiences. Even narrative possibilities not envisioned by the author can be improvised in response to open-ended player input, which significantly enhances player agency. However, while LLMs can effectively bridge authorial intent and player input ( Kim et al. 2024 ) , their use in IN also tends to widen the gap between the author’s envisioned narrative experience and the player’s actual experience. Using LLMs to interpret open-ended player input sacrifices the author’s comprehensive understanding of the range of actions players might perform. Moreover, using LLMs for just-in-time generation of narrative content restricts the author’s direct access to the specific narrative material that players may encounter ( Lu et al. 2025 ) . Consequently, authors of LLM-based IN struggle to anticipate the storylines that players may actually encounter, resulting in diminished control over player experience. This often leads to narratives that lack meaningful plot progression ( Tian et al. 2024 ; Chakrabarty et al. 2024 ; Beguš 2024 ) as the combination of player and LLM improvisation drags the storyline away from the intended course, and to failures of expressive communication ( Louie et al. 2022 ) as players lose touch with the human “hand behind the multiform plot” ( Murray 1997 , p. 347) . To address this challenge, we propose a novel concept of Bundled Storyline to aid authors of AI-based INs in making sense of narrative possibility space (Fig. 1 ). Our approach builds on insights from prior work in (AI-based) IN authoring ( Lu et al. 2025 ; Kreminski and Wardrip-Fruin 2018 ) , storyline visualization ( Chung et al. 2022 ; Masson et al. 2025 ) , and general-purpose tools for visualizing and making sense of possibility spaces ( Suh et al. 2024 ; Kreminski et al. 2022 ; Chung and Kreminski 2024 ; Smith and Whitehead 2010 ) . A bundled storyline structure “bundles together” narrative states that are indistinguishable in terms of certain user-configurable narrative dimensions to construct branching visualizations, which illustrate how different possible playthroughs of a single AI-based IN tend to converge and diverge in an abstract narrative state space. Narrative dimensions may be defined manually by an author (Fig. 1 (a)) to assess alignment between player experience and authorial intent, or automatically derived from player-experienced storylines to reveal emergent plot dynamics (Fig. 1 (b)). Additionally, users