Robust Preference Learning for Storytelling via Contrastive Reinforcement Learning
This addresses the challenge of controlled automated story generation for applications like creative writing or interactive systems, though it is incremental as it builds on existing contrastive and reinforcement learning techniques.
The paper tackled the problem of generating stories that satisfy user preferences by developing a pipeline that trains a contrastive reward model and fine-tunes a generative language model via reinforcement learning, resulting in a story generator preferred over a baseline LLM 20x larger and logit-based methods in a human study.
Controlled automated story generation seeks to generate natural language stories satisfying constraints from natural language critiques or preferences. Existing methods to control for story preference utilize prompt engineering which is labor intensive and often inconsistent. They may also use logit-manipulation methods which require annotated datasets to exist for the desired attributes. To address these issues, we first train a contrastive bi-encoder model to align stories with corresponding human critiques, named CARP, building a general purpose preference model. This is subsequently used as a reward function to fine-tune a generative language model via reinforcement learning. However, simply fine-tuning a generative language model with a contrastive reward model does not always reliably result in a story generation system capable of generating stories that meet user preferences. To increase story generation robustness we further fine-tune the contrastive reward model using a prompt-learning technique. A human participant study is then conducted comparing generations from our full system, ablations, and two baselines. We show that the full fine-tuning pipeline results in a story generator preferred over a LLM 20x as large as well as logit-based methods. This motivates the use of contrastive learning for general purpose human preference modeling.