Incorporating Global and Local Knowledge in Intentional Narrative Planning

The inclusion of independent, imperfect knowledge that represents virtual agents' belief of the local state of a narrative planning world has become a key component of narrative generation through simulation of multiple characters. However such models of belief incur significant computational cost. This paper demonstrates that despite the computational complexity, narratives can be generated not only as emergent stories in simulations, but also by global search using Planning that includes a model of differing, independent beliefs. We define a narrative state suitable for planning, detail how it incorporates belief, and how this can be used in an intent-based global search based planning algorithm. Two example narratives are used to illustrate how imperfect belief and social actions can be used in the generation process. The planning algorithm, which integrates global narrative planning with local character level belief reasoning, is fully implemented in a prototype system which was used in the experimental evaluation in which narratives were generated against several objective functions with both global and greedy search. The results show that intent-based planning with belief modelling is able to: generate narratives beyond the reach of planners that have complete knowledge; and also efficiently produce objectively higher quality narratives than those generated by evaluation of only local character knowledge and beliefs.