Storytelling with Adjustable Narrator Styles and Sentiments

Most storytelling systems to date rely on manually coded knowledge, the cost of which usually restricts such systems to operate within a few domains where knowledge has been engineered. Open Story Generation systems are capable of learning knowledge necessary for telling stories in a given domain. In this paper, we describe a technique that generates and communicates stories in language with diverse styles and sentiments based on automatically learned narrative knowledge. Diversity in storytelling style may facilitate different communicative goals and focalization in narratives. Our approach learns from large-scale data sets such as the Google N-Gram Corpus and Project Gutenberg books in addition to crowdsourced stories to instill storytelling agents with linguistic and social behavioral knowledge. A user study shows our algorithm strongly agrees with human judgment on the interestingness, conciseness, and sentiments of the generated stories and outperforms existing algorithms.