SC2 Computational Models of How People Make Stories
Humans are very good at perceiving events that have happened in the world and synthesizing them as stories that are easy to understand and interesting to follow. Indeed, such stories are one of the fundamental currencies for knowledge exchange in our culture, from literary master pieces to ordinary anecdotes in daily conversation. We are also capable of inventing new stories to act as vehicles of communication or tools for various goals, from learning to entertainment. Yet computers, which are currently our main recording devices, do not store knowledge in the form of narrative. Some parts of the human story making ability have been long-standing goals of AI -- such as the generation of simple stories -- and the fuller picture of how this ability operates has recently become the focus of considerable effort in the fields of narratology and computational creativity. These efforts are particularly relevant because unless we find a way to build computational models of the process of how humans construct stories, we face a future where our narrative understanding of the world will be side-lined by less natural representations that are easier to compute for machines.
The course addresses existing efforts at modelling the human story making ability from four different perspectives. First, it reviews existing theoretical efforts to understand stories in the light of how they might be understood computationally. Second, it visits some of the attempts to develop systems that can invent stories. Third, it describes ongoing efforts to compose stories automatically from conceptually described facts. Finally, it presents some of the insights into how human cognition is applied to the task, with a view to identifying clues that might help computational modelling.
1. Stories: what are they and/or why are they important to people?
2. Inventing stories: coming up with new stories to tell
3. Composing stories: learning to tell what you know as good stories
4. The process of story making: (guessing) how people go about doing it
The goal of the course is threefold:
1. to understand the importance of having computational models of story making
2. to explore the gaps between the narratological and the computational views of the task
3. to review existing efforts and outstanding challenges in the field
P. Gervás, “Computational Approaches to Storytelling and Creativity”, AI Magazine, Vol 30, No 3, 2009
P. Gervás, “Computational Drafting of Plot Structures for Russian Folk Tales”, Cognitive Computation, 2015.
P. Gervás, “Composing Narrative Discourse for Stories of Many Characters: a Case Study over a Chess Game”, Literary and Linguistic Computing, vol. 29, 2014.
P. Gervás, “An Exploratory Model of Remembering, Telling and Understanding Experience in Simple Agents”, in Proceedings of Computational Creativity, Concept Invention, and General Intelligence Workshop (C3GI 2016), Bolzano, Italy, 2016.
P. Gervás and León, C., “When Reflective Feedback Triggers Goal Revision: a Computational Model for Literary Creativity”, in AI and Feedback, IJCAI 2015, Buenos Aires, 2015.
HS 4Course requirements
Dr. Pablo Gervás of Universidad Complutense de Madrid is director of both the university’s NIL research group (nil.fdi.ucm.es) and the Instituto de Tecnología del Conocimiento (www.itc.ucm.es). He is one of the world’s leading experts on automatic generation of (fictional) stories and poetry, and has an extensive background in natural language generation, computational creativity and in narratology. His central research focus concerns the study of creativity as applied to the automated generation of literary artefacts with novelty, value and meaning for a human audience. His work on automated story generation has most recently contributed to the computer-assisted generation of a West-End musical in Britain titled “Beyond the Fence.”Website