From socio-emotional scenarios to expressive virtual narrators

Telling a story requires linking a series of significant events using statements to maintain tension, create suspense and allow time for the development of emotions. When automating this process, one major challenge is the generation of these filling sentences and ensuring that they are sufficiently consistent with the story. To this end we propose in this paper to use a knowledge representation model of a local coherent world. We present a scheme for automatic storytelling based on an ontological representation of concepts and natural language generation algorithms that dynamically build relations between concepts. Our algorithms scan the scenario to build a sentence skeleton that will be enriched using pseudo-random queries in the ontology. This allows us to enrich the story while keeping the storyline coherent. We end by discussing our evaluation and presenting our preliminary results.