Multi-Aspect Controllable Text Generation with Disentangled Counterfactual Augmentation

TL;DR

MAGIC is a novel method for multi-aspect controllable text generation that uses disentangled counterfactual augmentation to address attribute correlation imbalances, improving control accuracy.

Contribution

The paper introduces MAGIC, a new approach leveraging disentangled counterfactual augmentation to better handle attribute correlations in multi-aspect text generation.

Findings

Outperforms state-of-the-art baselines in various attribute correlation scenarios.

Effectively alleviates issues caused by imbalanced attribute correlations.

Enhances multi-aspect control through target-guided counterfactual augmentation.

Abstract

Multi-aspect controllable text generation aims to control the generated texts in attributes from multiple aspects (e.g., "positive" from sentiment and "sport" from topic). For ease of obtaining training samples, existing works neglect attribute correlations formed by the intertwining of different attributes. Particularly, the stereotype formed by imbalanced attribute correlations significantly affects multi-aspect control. In this paper, we propose MAGIC, a new multi-aspect controllable text generation method with disentangled counterfactual augmentation. We alleviate the issue of imbalanced attribute correlations during training using counterfactual feature vectors in the attribute latent space by disentanglement. During inference, we enhance attribute correlations by target-guided counterfactual augmentation to further improve multi-aspect control. Experiments show that MAGIC outperforms state-of-the-art baselines in both imbalanced and balanced attribute correlation scenarios. Our source code and data are available at https://github.com/nju-websoft/MAGIC.