Audience aware computational discourse generation for instruction and persuasion

• Main Author: Sayan, Eren Sila
• Other Authors: Patrick Henry Winston.
• Format: Others
• Language: English
• Published: Massachusetts Institute of Technology 2014
• Subjects: Electrical Engineering and Computer Science.
• Online Access: http://hdl.handle.net/1721.1/91868

Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014. === Cataloged from PDF version of thesis. === Includes bibliographical references (pages 131-134). === If we are to take artificial intelligence to the next level, we must further our understanding of human storytelling, arguably the most salient aspect of human intelligence. The idea that the study and understanding of human narrative capability can advance multiple fields, including artificial intelligence, isn't a new one. The following, however, is: I claim that the right way to study and understand storytelling is not through the traditional lens of human creativity, aesthetics or even as a plain planning problem, but through formulating storytelling as a question of goal driven social interaction. In particular, I claim that any theory of storytelling must account for the goals of the storyteller and the storyteller's audience. To take a step toward such an account, I offer a framework, which I call Audience Aware Narrative Generation, drawing inspiration in particular from narratology, cognitive science, and of course, computer science. I propose questions that we need to work on answering, and suggest some rudimentary starter thoughts to serve as guidelines for continued research. I picked a small subset of the proposed questions on which to focus my computational efforts: storytelling for teaching and persuasive storytelling. More specifically, I developed exploratory implementations for addressing this subset on the Genesis story understanding platform. The results have been encouraging: On the pedagogical side, my implementation models and simulates a teacher using the story of Macbeth to instruct a student about concepts such as murder, greed, and predecessor relationships in monarchies. On the persuasion side, my implementation models and simulates various different tellings of the classic fairy tale "Hansel and Gretel" so as to make The Witch appear likable in one, and unlikable in another; to make The Woodcutter appear to be a good parent just going through difficult times in one, and a bad parent in another. Perhaps the most amusing example however, especially in these days of sensationalized and highly subjective journalism, is that given a story of the cyber warfare between Russia and Estonia, my implementation can generate one telling of the story which makes Russia appear to be the aggressor, and yet another telling which makes Estonia appear to be the aggressor. And isn't that the story of history, politics, and journalism in one neat package! Overall, I have made four key contributions: I proposed Audience Aware Narrative Generation as a new framework for developing theories of storytelling; I identified important questions that must be answered by storytelling research and proposed initial plans of attack for them; I introduced storytelling functionality into the Genesis story understanding platform; and I implemented narrative discourse generators which produce a wide range of narratives, adapting accordingly to different audiences and goals. === by Eren Sila Sayan. === M. Eng.