Title:Learning to Generate Text in Arbitrary Writing Styles

Abstract:Prior work in style-controlled text generation has focused on tasks such as emulating the style of prolific literary authors, producing formal or informal text, and the degree of toxicity of generated text. Plentiful demonstrations of these styles are available, and as a result modern language models are often able to emulate them, either via prompting or discriminative control. However, in applications such as writing assistants, it is desirable for language models to produce text in an author-specific style on the basis of a small writing sample. We find that instruction-tuned language models can struggle to reproduce author-specific style demonstrated in a prompt. Instead, we propose to guide a language model to generate text in a target style using contrastively-trained representations that capture stylometric features. A central challenge in doing so is that an author's writing is characterized by surprising token choices under a generic language model. To reconcile this tension, we combine generative re-scoring to achieve an author-specific model, with discriminative control to ensure style consistency at the sequence-level. The combination of these approaches is found to be particularly effective at adhering to an author-specific style in a variety of conditions, including unconditional generation and style transfer, and is applicable to any underlying language model without requiring fine-tuning.