Open Domain Collaborative Storytelling with Say Anything

In this demonstration we present Say Anything, an open domain interactive storytelling application where an author's original story sentences are used to select subsequent sentences from a corpus of millions of stories extracted from Internet weblogs. Demonstration: Say Anything Interactive storytelling has become a rich area of research for computer science, interactive-media and the learning sciences. In this demonstration we will present our system, Say Anything (Swanson and Gordon 2008). This application uses a different approach from most interactive storytelling systems in that it takes advantage of massive amounts of weblog text to enable completely open domain story generation. Say Anything breaks from recent interactive storytelling systems in its interaction model as well as its underlying architecture. Unlike most current systems, which use rich immersive 3D virtual environments, the interaction with our system is purely textual. Although 3D virtual environments offer many advantages and possibilities that were not possible when research in this area first began, there are still many reasons that textual environments have significant value (Montfort 2007). In our system, a human and computer interact by taking turns contributing individual sentences of an emerging story. The human user begins the story with the first sentence, which is continued by the computer’s response. This process continues cyclically until the user is satisfied (or sufficiently dissatisfied) with their story. The computer performs three major operations to generate a sentence for the developing story. First, it analyzes lexical and grammatical features of the user’s story, including the most recently contributed sentence. This analysis is then used to retrieve the most similar sentence from a database of millions of stories collected from Internet weblogs (Gordon et al. 2007). The measure of similarity is based on standard information retrieval techniques implemented in the Apache Lucene search Copyright © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. engine (Gospodnetic and Hatcher 2004). The system then selects the next sentence in the story where the retrieved sentence appears. In addition to this sentence, the user is also presented with nine other sentences that follow from the next most similar sentences, which they can choose from using a drop-down menu. Alternatively, if none of the sentences are satisfactory to the user, they may opt to remove it completely and continue the story by writing another sentence themselves. On the other hand if the user is pleased with the direction the computer is taking the story they can allow the computer to continue generating sentences, one at a time. See Figure 1 for an example story written by one of our users. Our system is collaborative in the sense that the user and computer are working together to create a single story. It is also collaborative in the sense that our database consists of stories from hundreds of thousands of weblog authors who contribute a small piece to each new story created by our system. Keeping with this collaborative spirit we chose to use a web interface for our system in hopes of spurring a community that can write, share and rate each other’s stories. See Figure 2 for an illustration of the main writing interface page of our system. In the demonstration of our system we will allow users to author their own original stories in real time. In addition, the users will be able to explore the features we have implemented to try to foster a web-based community around this type of simple storytelling game.