Controllable Text Generation for Large Language Models: A Survey

TL;DR

This survey defines controllable text generation (CTG) for large language models, clarifying two dimensions of control (content hard control and attribute soft control) and balancing control with text quality such as fluency and diversity. It catalogs training-stage and inference-stage methods, including retraining, fine-tuning, RL, prompt engineering, latent-space manipulation, and decoding-time interventions, with emphasis on Transformer-based LLMs. Key contributions include a unified task classification, a comprehensive method taxonomy, evaluation frameworks, and practical domain and general-task applications, along with challenges like multi-attribute decoupling and decoding-time efficiency. The work highlights real-world relevance and calls for more application-oriented research, broader testing scenarios, and improved baselines to harness LLM capabilities while ensuring safe, useful generation.

Abstract

In Natural Language Processing (NLP), Large Language Models (LLMs) have demonstrated high text generation quality. However, in real-world applications, LLMs must meet increasingly complex requirements. Beyond avoiding misleading or inappropriate content, LLMs are also expected to cater to specific user needs, such as imitating particular writing styles or generating text with poetic richness. These varied demands have driven the development of Controllable Text Generation (CTG) techniques, which ensure that outputs adhere to predefined control conditions--such as safety, sentiment, thematic consistency, and linguistic style--while maintaining high standards of helpfulness, fluency, and diversity. This paper systematically reviews the latest advancements in CTG for LLMs, offering a comprehensive definition of its core concepts and clarifying the requirements for control conditions and text quality. We categorize CTG tasks into two primary types: content control and attribute control. The key methods are discussed, including model retraining, fine-tuning, reinforcement learning, prompt engineering, latent space manipulation, and decoding-time intervention. We analyze each method's characteristics, advantages, and limitations, providing nuanced insights for achieving generation control. Additionally, we review CTG evaluation methods, summarize its applications across domains, and address key challenges in current research, including reduced fluency and practicality. We also propose several appeals, such as placing greater emphasis on real-world applications in future research. This paper aims to offer valuable guidance to researchers and developers in the field. Our reference list and Chinese version are open-sourced at https://github.com/IAAR-Shanghai/CTGSurvey.

Figures (11)

• Figure 1: Publication trends on Web of Science related to Controllable Generation in Language Models.
• Figure 2: Controllability dimension and capability dimension of LLMs.
• Figure 3: Survey Framework
• Figure 4: Injection of Conditions in CTG
• Figure 5: Classification of Controllable Text Generation Methods