First Person Narrative Story Extraction and Retrieval

The ability to acquire, communicate and transfer complex knowledge is one of the fundamental characteristics that make us human. It should be no surprise that knowledge representation and management is seen as a fundamental problem in Artificial Intelligence research. There is certainly no requirement that a machine encode this knowledge the same way humans do. However, any decisions in representation must be made with great care if there is any hope of computer agents being able to communicate effectively and coherently in a human environment. Like many other problems finding the right level of granularity for representation is a difficult open problem. Schank & Abelson [18] argue that, despite longstanding beliefs, propositional knowledge is not the primary mechanism for knowledge storage or communication in humans. Rather stories, the first person narrative accounts of events, are much more fundamental and offer a more flexible structure for manipulation. For example, stories about similar events share many structural elements often referred to as the script or skeleton. Slight perturbations in the telling of a story can be used to embellish the same story to convey different moral outcomes or elicit different emotional responses from the listener. Multiple views are also possible from a single story offering different interpretations based on the role of the teller. The ability to pass knowledge through the telling of such a rich representation makes the value of a large collection of stories immediately obvious. In fact story telling is seen as an important process in the management sciences for assessing performance and directions for change. Another important application, particularly useful to the projects at the Institute for Creative Technologies, is the integration of story collections into training technologies and simulations. Johnson et al. [9] and Gordon [3] have separately shown ways to integrate story collections into a working environment to improve the training and performance of individuals. Despite the inescapable prevalence of stories around us an effective means for collecting and managing large-scale repositories has yet to be developed. Several manual collection programs have been tried, in specialized domains, such as the WPA Life Histories Collection by the Library of Congress (memory.loc.gov/ammem/wpaintro/). Unfortunately large-scale manual collection endeavors such as these are astronomically expensive and require an extraordinary amount of time to put together. Another collection strategy is through a guided interview process where a trained expert stimulates the conversation in a way to maximize the story content told. If non-fictional stories are not a requirement then a third possible way to acquire a collection of stories is simply to hire a staff of writers. This approach, too, is impractical for very large collections and often lacks the breadth of details that are found in real world experiences. Every year sees a vast increase in the number of people with online access. This incredible rise of users has created a critical mass that has materialized into large online communities and an explosion of personal content being exchanged through the popularity of Weblogs. With over 70 million accessible Weblogs vast amounts of stories are