Enhancing story generation with the semantic web
Eric LaBouve, Erik Miller, Foaad Khosmood · 2019
In story or character driven games, in-game stories are usually manually authored in advance. As the complexity of interactions in games increases, the quantity of hand-crafted text typically follows. Designing stories and composing content by hand is a laborious and time consuming process that if automated, would speed up game production and lower development costs. In this paper, we present a mixed initiative tool to help generalize and enhance context free grammars (CFGs) for story generation. The tool is designed to take as input a story generating grammar in addition to generic keywords for people, places and other various metrics in order to control the output text. The tool is knowledgeable about a wide array of topics because it leverages the Semantic Web in order to extrapolate more details and related information from the user supplied content. As a result, generated text will contain genuinely new information, descriptions of characters and locations that were never written by the author. Although the general structure of a story or dialogue is somewhat fixed by the nature of grammar rules, the resulting text can be geared towards a variety of user inputs and can include details that may surprise designers. The tool is evaluated by a group of 15 individuals in a user study to gauge the practicality of using Semantic Web technologies for procedural text generation. The study concludes that using the Semantic Web is an effective aid for grammar based text generation. We discuss our system, the user study and share thoughts on future work.