Capturing Classic Authorial Style in Long-Form Story Generation with GRPO Fine-Tuning

TL;DR

This paper introduces a two-stage method using a style-similarity judge and GRPO fine-tuning to improve authorial style in long-form story generation, achieving higher style scores across multiple authors.

Contribution

It presents a novel approach combining a style judge with GRPO to enable controllable style transfer without accept/reject supervision.

Findings

The GRPO-trained model scores higher on style than baselines.

Style scores averaged 0.893 across four authors.

AV-calibrated reward modeling effectively controls style in generation.

Abstract

Evaluating and optimising authorial style in long-form story generation remains challenging because style is often assessed with ad hoc prompting and is frequently conflated with overall writing quality. We propose a two-stage pipeline. First, we train a dedicated style-similarity judge by fine-tuning a sentence-transformer with authorship-verification supervision, and calibrate its similarity outputs into a bounded $[0,1]$ reward. Second, we use this judge as the primary reward in Group Relative Policy Optimization (GRPO) to fine-tune an 8B story generator for style-conditioned writing, avoiding the accept/reject supervision required by Direct Preference Optimization (DPO). Across four target authors (Mark Twain, Jane Austen, Charles Dickens, Thomas Hardy), the GRPO-trained 8B model achieves higher style scores than open-weight baselines, with an average style score of 0.893 across authors. These results suggest that AV-calibrated reward modelling provides a practical mechanism for controllable style transfer in long-form generation under a moderate model size and training budget.