Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes

TL;DR

Plug-and-Blend is a controllable story generation framework that enables users to influence story topics and transitions, including blended topics, using multiple control codes with maintained fluency and coherence.

Contribution

It introduces a novel plug-and-play framework allowing fine-grained and blended topic control in story generation with pre-trained language models.

Findings

Effective control of story topics using control codes.

Strong blending capability between multiple topics.

Generated stories maintain fluency and coherence.

Abstract

Large pre-trained neural language models (LM) have very powerful text generation capabilities. However, in practice, they are hard to control for creative purposes. We describe a Plug-and-Play controllable language generation framework, Plug-and-Blend, that allows a human user to input multiple control codes (topics). In the context of automated story generation, this allows a human user loose or fine-grained control of the topics and transitions between them that will appear in the generated story, and can even allow for overlapping, blended topics. Automated evaluations show our framework, working with different generative LMs, controls the generation towards given continuous-weighted control codes while keeping the generated sentences fluent, demonstrating strong blending capability. A human participant evaluation shows that the generated stories are observably transitioning between two topics.