Multi-Aspect Controllable Text Generation with Disentangled Counterfactual Augmentation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917679335473152 |
|---|---|
| author | Liu, Yi Liu, Xiangyu Zhu, Xiangrong Hu, Wei |
| author_facet | Liu, Yi Liu, Xiangyu Zhu, Xiangrong Hu, Wei |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_19958 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Multi-Aspect Controllable Text Generation with Disentangled Counterfactual Augmentation Liu, Yi Liu, Xiangyu Zhu, Xiangrong Hu, Wei Computation and Language Artificial Intelligence 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. |
| title | Multi-Aspect Controllable Text Generation with Disentangled Counterfactual Augmentation |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2405.19958 |