Modeling the Internal Monologue: A Framework for Reconciling Authorial Intent with Agent Autonomy

Abstract

The design of narrative agents is often caught in a trade-off between authorial control and character agency. This work proposes the existence of a middle-ground approach, arguing that a robust, authored narrative can emerge from the performative autonomy of deeply simulated, coherent agents. To explore this thesis, I have developed a novel, three-layer architecture that synthesizes Stanislavskian performance theory, defeasible logic for argumentation-based reasoning, and a dynamic behavior language for execution. My work to date includes implementing this architecture for goal attainment in the Oops! All Bards prototype, and exploring foundational personality models for goal formation in the Stay Thy Blade prototype. The remaining research directions presented for feedback include the synthesis of these two tracks, the development of a constrained knowledge authoring framework, and a multi-pronged evaluation strategy. The contribution of this dissertation will be a validated framework that attempts to bridge the gap between agent autonomy and authorial intent, enabling the creation of "strong stories through strong characters."