Declarative Optimization-Based Drama Management (DODM) in the Interactive Fiction Anchorhead

A drama manager guides a player through a story experience by modifying the experience in reaction to the player’s actions. Declarative optimization-based drama management (DODM) casts the drama-management problem as an optimization problem: The author declaratively specifies a set of plot points in a story, a set of actions the drama manager can take, and an evaluation function that rates a particular story. The drama manager then takes the actions in a way that attempts to maximize story quality. Peter Weyhrauch reported good results using a variant of game-tree search to optimize the use of drama-manager actions. We attempt to replicate these results on another story, Anchorhead, and show that search does not perform very well in general, especially on larger and more complex stories. However, we believe that this is a problem with the specific optimization method, not the general approach, and report some results demonstrating the plausibility of applying reinforcement-learning techniques to compute a policy instead of search.