Controllable Text Generation with Focused Variation

TL;DR

This paper presents the Focused-Variation Network (FVN), a new model for controllable text generation that enhances attribute control and diversity while maintaining fluency, achieving state-of-the-art results.

Contribution

FVN introduces disjoint discrete latent spaces for each attribute, improving controllability and diversity in generated text compared to prior models.

Findings

FVN achieves state-of-the-art performance on annotated datasets.

FVN generates diverse and attribute-controlled fluent text.

Automatic and human evaluations confirm FVN's effectiveness.

Abstract

This work introduces Focused-Variation Network (FVN), a novel model to control language generation. The main problems in previous controlled language generation models range from the difficulty of generating text according to the given attributes, to the lack of diversity of the generated texts. FVN addresses these issues by learning disjoint discrete latent spaces for each attribute inside codebooks, which allows for both controllability and diversity, while at the same time generating fluent text. We evaluate FVN on two text generation datasets with annotated content and style, and show state-of-the-art performance as assessed by automatic and human evaluations.