Narrative Planning: Balancing Plot and Character

Narrative, and in particular storytelling, is an important part of the human experience. Consequently, computational systems that can reason about narrative can be more effective communicators, entertainers, educators, and trainers. One of the central challenges in computational narrative reasoning is narrative generation, the automated creation of meaningful event sequences. There are many factors – logical and aesthetic – that contribute to the success of a narrative artifact. Central to this success is its understandability. We argue that the following two attributes of narratives are universal: (a) the logical causal progression of plot, and (b) character believability. Character believability is the perception by the audience that the actions performed by characters do not negatively impact the audiences suspension of disbelief. Specifically, characters must be perceived by the audience to be intentional agents. In this article, we explore the use of refinement search as a technique for solving the narrative generation problem – to find a sound and believable sequence of character actions that transforms an initial world state into a world state in which goal propositions hold. We describe a novel refinement search planning algorithm – the Intent-based Partial Order Causal Link (IPOCL) planner – that, in addition to creating causally sound plot progression, reasons about character intentionality by identifying possible character goals that explain their actions and creating plan structures that explain why those characters commit to their goals. We present the results of an empirical evaluation that demonstrates that narrative plans generated by the IPOCL algorithm support audience comprehension of character intentions better than plans generated by conventional partial-order planners.

💡 Research Summary

The paper addresses the longstanding challenge of automatically generating coherent narratives that are both logically sound and emotionally believable. It begins by asserting that two universal attributes underlie successful stories: (a) a causal progression of plot events that transforms an initial world state into a goal state, and (b) character believability, which requires that audiences perceive characters as intentional agents whose actions do not break suspension of disbelief. While conventional planning approaches—particularly Partial Order Causal Link (POCL) planners—excel at constructing causally valid event sequences, they lack mechanisms for explaining why a character chooses a given action, leaving audiences to infer intentions on their own.

To bridge this gap, the authors introduce the Intent‑based Partial Order Causal Link (IPOCL) planner, a refinement‑search algorithm that augments the classic POCL framework with three key innovations. First, IPOCL performs a backward inference of possible character goals for each action, using the pre‑ and post‑conditions of the action together with the surrounding causal network. This step yields a set of candidate goals that could have motivated the observed behavior. Second, the planner adds “intentional links” that explicitly connect inferred goals to the actions that serve them, thereby constructing a hierarchical goal‑network where higher‑level goals support lower‑level ones. Third, during the refinement phase, IPOCL simultaneously checks for causal consistency (no unsatisfied preconditions, no threats to existing links) and intentional consistency (no contradictions between inferred goals and the overall goal hierarchy). If a conflict is detected, the planner either modifies the offending action, introduces additional supporting actions, or revises the goal assignments until both dimensions are satisfied.

The algorithm proceeds as follows: (1) define the initial world state and the set of story‑ending propositions; (2) generate a provisional POCL plan that achieves the story goals; (3) for each action, infer plausible character goals; (4) insert intentional links to create a combined causal‑intentional graph; (5) iteratively refine the plan, resolving any causal threats or intentional inconsistencies; (6) output a final narrative plan where every action is both causally justified and intentional‑ly explained.

Empirical evaluation involved two conditions. In the control condition, participants read narratives produced by a standard POCL planner. In the experimental condition, participants read narratives generated by IPOCL. After reading, participants answered questionnaires measuring their comprehension of character intentions, perceived coherence, and overall enjoyment. Statistical analysis showed that IPOCL‑generated stories yielded significantly higher intention‑understanding scores, confirming that the explicit intentional structure aids audience comprehension without sacrificing plot coherence.

The contributions of the paper are threefold. (1) It formalizes the dual‑objective narrative generation problem, integrating logical plot progression with character intentionality. (2) It introduces intentional links as a novel planning construct, enabling the representation of multi‑layered character goal hierarchies within a partial‑order planning framework. (3) It provides an empirical validation that the combined causal‑intentional plans improve human readers’ mental models of character agency.

Beyond the immediate results, IPOCL’s modular design allows it to be combined with existing POCL systems, making it adaptable to a variety of interactive media such as role‑playing games, educational simulations, and adaptive storytelling platforms. Future work suggested by the authors includes refining the goal‑inference mechanism (e.g., incorporating affective motivations), extending the model to handle dynamic audience feedback in real time, and scaling the approach to larger, more complex story worlds. In sum, the paper demonstrates that a refinement‑search planner that simultaneously reasons about plot causality and character intentionality can produce narratives that are both logically sound and more readily understood by human audiences.