Narrative Interpolation for Generating and Understanding Stories
This addresses the problem of generating structured and coherent stories for applications like creative writing or interactive storytelling, though it is incremental as it builds on existing language models like GPT-2.
The paper tackles controlled narrative generation by guiding a model to produce coherent stories with user-specified endings through interpolation, showing that this approach yields narratives that are coherent, faithful to the ending, and reduce manual effort compared to past methods.
We propose a method for controlled narrative/story generation where we are able to guide the model to produce coherent narratives with user-specified target endings by interpolation: for example, we are told that Jim went hiking and at the end Jim needed to be rescued, and we want the model to incrementally generate steps along the way. The core of our method is an interpolation model based on GPT-2 which conditions on a previous sentence and a next sentence in a narrative and fills in the gap. Additionally, a reranker helps control for coherence of the generated text. With human evaluation, we show that ending-guided generation results in narratives which are coherent, faithful to the given ending guide, and require less manual effort on the part of the human guide writer than past approaches.