To construct believable flow and nuanced dialogues in a BDI framework: An architecture for the development of interactive narratives

Abstract

There are many games with narrative aspects in which the player will have to speak with non-playable characters (NPC). These games exist both in the commercial gaming industry and the serious gaming industry. Serious games are games that are created to train or educate. The dialogues that take place in those games are often well-written. Yet, the characters are seldom open for much interaction. The reason for this, is that the dialogues in games are most often fully scripted and interaction would quickly cause a combinatorial explosion of possible storylines. We would like to increase interaction with the NPC’s in serious games while still retaining some of the expressive control currently provided by scripting dialogues, because we believe it will offer new opportunities for training and education. Especially for games that educate on bad news reports, sales tactics, interrogating witnesses or suspects, and other similar situations in which eloquence and an understanding of the nuance of human natural language are key to success. Current approaches for increasing the interactivity of NPC’s do not offer the artists enough expressive control over the agents for this kind of games to be used as business product. We argue that it is essential for an interactive character to be believable and social and that those aspects should permeate any architecture designed for the realisation of such an agent. That is why we propose to combine a Belief, Desire, Intention framework (BDI), with scripting. BDI offers us generality and modularity. Scripting offers us expressive control, because we can author specific parts of a dialogue by hand. The agent’s action-selection mechanism is a partial implementation of JADEX (a BDI reasoning engine) into ActionScript. The agent's action-expression mechanism consists of two tools, a tool called OZ Dialog Editor, in which scripts can be authored, and a tool called OZ Dialog Player, which can run the scripts. The agent selects plans based on its beliefs, desires, and intentions. The execution of a plan consists of at least one action. An action opens a script as the expression of that action. The architecture we designed is a first step towards creating believable and nuanced flow in a BDI framework. In game studios, programmers program the NPC’s and game designers write the dialogues. They can do this separately. We aim to sustain this task-segregation between programming agents and writing dialogues. This task-segregation bears restrictions that shape our architecture. The architecture was tested with two game designers at &ranj serious games by paper-prototyping. The game designers performed an exercise in which they had to use certain script building blocks to write scripts. After this exercise we conducted an interview where we discussed the functionality of the building block and the architecture in general. The shape and function of the building block was adjusted based on the results of the interview. In general, the tests indicated that our method to creating interactive narratives could work. However, we argue that there is a large gap between theory and practice in AI research. Often issues will only be addressed if the method is actually put to work. Therefore, we suggest future researchers to build a game using our architecture.