# Control, Generate, Augment: A Scalable Framework for Multi-Attribute Text Generation

arXiv: https://arxiv.org/abs/2004.14983 · HTML: https://mlsanity.com/paper/2004.14983/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 30 Apr 2020.

## Built on by (most cited first, 18 of 18)

- [Large Language Model as Attributed Training Data Generator: A Tale of Diversity and Bias](https://mlsanity.com/paper/2306.15895/) (2023)
- [DoCoGen: Domain Counterfactual Generation for Low Resource Domain Adaptation](https://mlsanity.com/paper/2202.12350/) (2022)
- [FlipDA: Effective and Robust Data Augmentation for Few-Shot Learning](https://mlsanity.com/paper/2108.06332/) (2021)
- [Personalized Text Generation with Fine-Grained Linguistic Control](https://mlsanity.com/paper/2402.04914/) (2024)
- [Valid Survey Simulations with Limited Human Data: The Roles of Prompting, Fine-Tuning, and Rectification](https://mlsanity.com/paper/2510.11408/) (2025)
- [Spillover of Antisocial Behavior from Fringe Platforms: The Unintended Consequences of Community Banning](https://mlsanity.com/paper/2209.09803/) (2022)
- [MACSum: Controllable Summarization with Mixed Attributes](https://mlsanity.com/paper/2211.05041/) (2022)
- [Boosting Reward Model with Preference-Conditional Multi-Aspect Synthetic Data Generation](https://mlsanity.com/paper/2407.16008/) (2024)
- [Understanding Online Migration Decisions Following the Banning of Radical Communities](https://mlsanity.com/paper/2212.04765/) (2022)
- [Controllable Dialogue Generation with Disentangled Multi-grained Style Specification and Attribute Consistency Reward](https://mlsanity.com/paper/2109.06717/) (2021)
- [Stranger Danger! Cross-Community Interactions with Fringe Users Increase the Growth of Fringe Communities on Reddit](https://mlsanity.com/paper/2310.12186/) (2023)
- [Attributes as Textual Genes: Leveraging LLMs as Genetic Algorithm Simulators for Conditional Synthetic Data Generation](https://mlsanity.com/paper/2509.02040/) (2025)
- [Learning from Multiple Noisy Augmented Data Sets for Better Cross-Lingual Spoken Language Understanding](https://mlsanity.com/paper/2109.01583/) (2021)
- [Disentangling Active and Passive Cosponsorship in the U.S. Congress](https://mlsanity.com/paper/2205.09674/) (2022)
- [Dynamic Human Evaluation for Relative Model Comparisons](https://mlsanity.com/paper/2112.08048/) (2021)
- [Bootstrapping Relation Extractors using Syntactic Search by Examples](https://mlsanity.com/paper/2102.05007/) (2021)
- [UPB @ ACTI: Detecting Conspiracies using fine tuned Sentence Transformers](https://mlsanity.com/paper/2309.16275/) (2023)
- [Controlled Attribute-Specific Summarization of Interrogative Dialogues](https://mlsanity.com/paper/2609.28004/) (2026)

## Abstract (arXiv)

We introduce CGA, a conditional VAE architecture, to control, generate, and augment text. CGA is able to generate natural English sentences controlling multiple semantic and syntactic attributes by combining adversarial learning with a context-aware loss and a cyclical word dropout routine. We demonstrate the value of the individual model components in an ablation study. The scalability of our approach is ensured through a single discriminator, independently of the number of attributes. We show high quality, diversity and attribute control in the generated sentences through a series of automatic and human assessments. As the main application of our work, we test the potential of this new NLG model in a data augmentation scenario. In a downstream NLP task, the sentences generated by our CGA model show significant improvements over a strong baseline, and a classification performance often comparable to adding same amount of additional real data.