The primary objective of this research is to develop new cognitively informed plan-based models of narrative action and to demonstrate that these models can be used both to control a virtual environment and to make effective predictions about the results of users' mental models of the stories that they characterize. Motivated by psychological models of plans and plan reasoning, this research builds on prior work in plan generation and plan-related communication to develop an architecture for creating understandable interaction in narrative-oriented virtual environments. The specific research program can be divided into two high-level thrusts: 1) Developing new generative knowledge representation schemes for the control of narrative action, focusing on the structures of conflict and goal dynamics. 2) Formally validating the results from the items via large-scale empirical evaluations.
This work will develop computational models of narrative, focusing on elements of creativity in narrative (as defined roughly by coherence and expectation violation). The project will explore the hypothesis that creativity in the design of many artifacts (and in the design of narrative in particular) is not only a property of the algorithms used to create the artifacts but also a property of how the artifacts are experienced or understood by human users.
This work will have a significant impact on the theory and understanding of the relationships between computation and cognition, particularly in the context of narrative. Because of the multidisciplinary nature of the research objectives, the project will produce significant advances in both computer science and cognitive science. It is anticipated that the resulting model will serve as a foundation for a new generation of tools that support mixed-initiative virtual world design, particularly focusing on the generation of narrative systems. In addition, the research will explore the use of the models to create customized, context-sensitive storylines for computer game-based learning environments.
The project will contribute to the infrastructure of science and education by training new researchers (graduate research assistants) in an area that is broadly multidisciplinary (computer science, cognitive science and narrative theory). These new researchers will gain from the project a unique integrated view of the contributing disciplines. The project will train undergraduates through involvement in formal and informal research exposure efforts supported in part by REU supplements.
Narrative plays a central role in both our understanding of the world around us as well as in the ways that we communicate with one another about our own and others' experiences. This project has worked to develop some of the first computational models of key aspects of narrative. Further, the researchers have leveraged those models generatively, as components of systems that create narratives automatically. Finally, the project has provided experimental validation of the techniques, showing that human subjects reading or experiencing the automatically created narratives percieve the stories has structured in ways our models predict. One branch of the project was devoted to the design, implementation, and evaluation of a computational model of conflict in narrative. Conflict is an essential element of interesting stories because it provides motivation for the characters and structure for the audience. The model developed in this project is meant to be used by automatic story writing systems to make the stories they generate more interesting and more similar to stories written by humans. Many story generation systems use an Artificial Intelligence technique called planning as their underlying algorithm. Early planners, which were designed for tasks like robot movement and factory scheduling, reasoned explicitly about conflict so that all conflict could be removed from a plan. The story planner CPOCL, developed by this project, uses this same model of conflict, but rather than eliminating conflict, it seeks to create conflict that is interesting according to certain metrics.??The model is dubbed "the CPOCL model of conflict," and it has been evaluated in two empirical experiments to date. The first demonstrated that humans who read stories generated by the CPOCL planner can identify which conflicts exists, which characters are in conflict, and how long each conflict lasts. The second demonstrated that the CPOCL formulas for measuring metrics like the intensity and balanced of a conflict reflected the perception of human readers. ??The second branch of the research developed the Intention Revision in Storytelling (IRIS) system, a narrative generation tool that creates story outlines with built-in suspense. Like the work we undertook on narrative conflict, this work also builds on AI planning methods, creating suspense by introducing plan failure into the protagonist's plan and having the protagonist perform intention revision to adapt to the resulting new circumstances. The algorithmic components used to create suspense were experimentally validated as part of this effort. The development of computational models of intention revision, suspense and conflict in stories is of most direct benefit in the area of automatic story generation. Currently, systems that provide narrative experiences to engage, educate or entertain currently make use of pre-authored, static narratives. As a result, experiences in these systems are constrained to a one-size-fits-all approach. The models developed in this research project enable new types of systems that can adapt their stories to the actions, goals and learning objectives of each user on the fly. This capability will enable applications like automatic tutoring systems and training simulations to benefit by increasing their ability to automatically generate stories which meet the narrative expectations of the audience and are tailored to individual needs at run-time. Beyond the technical and scientific outcomes of the work, the project had significant impact on the training of both computer science undergraduates and graduate students. The project supported a number of undergraduates through supplements targeting research experiences for undergraduates, providing both advanced training in 3D virtual world development as well as exposure to research methods and activities in artificial intelligence. The project also supported the doctoral studies of two computer science Ph.D. students that are both now in the final semesters of their academic programs.
