Reproduce “Automated Story Generation as Question-Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Louis Castricato, Spencer Frazier, Jonathan Balloch, Nitya Tarakad, Mark Riedl
EDGAR (Explanatory Drama Generation And Recall) is a system for automated story generation that works backward from a given ending using commonsense reasoning and question-answering to create coherent narratives. The system achieved a 15.9% improvement in objective coherence (Fabula Entropy) over neural baselines and was subjectively preferred twice as often by human evaluators.
Key insights
- Effective storytelling fundamentally involves explanation, where each event answers why subsequent events occur, suggesting that generating stories backward can ensure goal-orientation and strong causal coherence.
- Integrating multiple specialized neural models (commonsense inference, question-answering, and perplexity-based ranking) can address complex creative tasks like story generation more effectively than single auto-regressive models.
- Leveraging commonsense inferences to generate explanatory questions provides a structured way to guide story creation and maintain coherence toward a predetermined ending.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.