Say Anything: Learning to Tell Stories by Learning and Interacting
| « Previous Event | Next Event » |
Reid Swanson
Monday, November 7, 2011, 1:00 PM to 2:00 PM
Location: Engineering 2, Room 506
Hosted By the Center for Games and Playable Media
Abstract:
Digital Interactive Storytelling is a compelling medium for expressing ideas that transforms a normally passive experience into an active engagement in the creative process. Despite its enormous potential, the cost associated with authoring interactive narratives that provide choices the participants find satisfying has been a severe bottleneck to their widespread adoption. In this talk, I will present a text-based interactive storytelling game that addresses this problem by leveraging the millions of freely available stories written by ordinary people. These stories provide the backbone of a large-scale case-based reasoning system that drives the narrative generation. I will show that even a simple architecture using only a single sentence of context and standard information retrieval techniques works surprisingly well. However, by using arbitrary features of the text and gameplay interaction as training data for machine learning, we can produce complete story traces that are of similar quality to entirely human authored stories drawn from our (held-out) weblog corpus.
Bio:
Reid Swanson has a diverse multi-disciplinary background that includes Fine Art, Philosophy, Linguistics and Computer Science. In 2010, he received his Ph.D. in Computer Science from the University of Southern California while working as a GRA with his advisor Andrew Gordon at the Institute for Creative Technologies. His research at USC focused on using large-scale corpora and textual knowledge representations to enable open-domain reasoning for interactive storytelling. After graduating, Reid spent a year at Walt Disney Imagineering’s Research & Development lab where he worked on bringing interactive stories to their theme parks and resorts. Currently Reid is working as a Postdoc here at UCSC with Arnav Jhala in the Computational Cinema group.
