Joao G. M. Araujo | Searcharxiv

arXiv2026

Fabula: Building a Narrative Storytelling Sidekick with the Writers' Community

We design and evaluate Fabula, an interactive app for fiction writers. Fabula uses detailed narrative plans informed by general narratological theory. Stories are structured hierarchically into scenes and beats that can be (re)generated and revised at script and story plan level. Using participatory AI, we critically evaluate and improve Fabula with casual and published writers, via design interviews and writing sessions with 42 experts, and large-scale internal and external testing. We interrogate our design choices: (1) whether a language model-based auto-evaluator, optimized on human experts' preferences, can improve story quality, (2) whether users want UI that exposes the detailed narrative plan alongside the story script, (3) to what extent our narratology assumptions fit localised storytelling traditions and serve screenwriters or playwrights, and (4) whether convergent iteration over the story plan supports writers' creativity. Building on critical feedback and concerns, we use Fabula as a cultural probe in adversarial design, and identify potentials for writing feedback and for interactive storytelling.

cs.HC↗

arXiv2025

Wasserstein Policy Optimization

We introduce Wasserstein Policy Optimization (WPO), an actor-critic algorithm for reinforcement learning in continuous action spaces. WPO can be derived as an approximation to Wasserstein gradient flow over the space of all policies projected into a finite-dimensional parameter space (e.g., the weights of a neural network), leading to a simple and completely general closed-form update. The resulting algorithm combines many properties of deterministic and classic policy gradient methods. Like deterministic policy gradients, it exploits knowledge of the gradient of the action-value function with respect to the action. Like classic policy gradients, it can be applied to stochastic policies with arbitrary distributions over actions -- without using the reparameterization trick. We show results on the DeepMind Control Suite and a magnetic confinement fusion task which compare favorably with state-of-the-art continuous control methods.

cs.LG↗