# Natural Language Actor-Critic: Scalable Off-Policy Learning in Language Space

arXiv: https://arxiv.org/abs/2512.04601 · HTML: https://mlsanity.com/paper/2512.04601/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods.
- Builds on 30 of its 48 references that are in the map.
- First posted on arXiv 4 Dec 2025.

## Placed in

- Deep reinforcement learning methods / [RL training of language and agent models](https://mlsanity.com/deeprl/rl-training-of-language-and-agent-models/)

## Builds on (most cited first, 30 of 30)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Toolformer: Language Models Can Teach Themselves to Use Tools](https://mlsanity.com/paper/2302.04761/) (2023)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Learning without Forgetting](https://mlsanity.com/paper/1606.09282/) (2016)
- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https://mlsanity.com/paper/2305.16291/) (2023)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [Prioritized Experience Replay](https://mlsanity.com/paper/1511.05952/) (2015)
- [A Distributional Perspective on Reinforcement Learning](https://mlsanity.com/paper/1707.06887/) (2017)
- [Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models](https://mlsanity.com/paper/2310.04406/) (2023)
- [Training Language Models to Self-Correct via Reinforcement Learning](https://mlsanity.com/paper/2409.12917/) (2024)
- [Successor Features for Transfer in Reinforcement Learning](https://mlsanity.com/paper/1606.05312/) (2016)
- [Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning](https://mlsanity.com/paper/2410.08146/) (2024)
- [Is Reinforcement Learning (Not) for Natural Language Processing: Benchmarks, Baselines, and Building Blocks for Natural Language Policy Optimization](https://mlsanity.com/paper/2210.01241/) (2022)
- [ArCHer: Training Language Model Agents via Hierarchical Multi-Turn RL](https://mlsanity.com/paper/2402.19446/) (2024)
- [Grounding Large Language Models in Interactive Environments with Online Reinforcement Learning](https://mlsanity.com/paper/2302.02662/) (2023)
- [Q-Transformer: Scalable Offline Reinforcement Learning via Autoregressive Q-Functions](https://mlsanity.com/paper/2309.10150/) (2023)
- [Prompt-Based Monte-Carlo Tree Search for Goal-Oriented Dialogue Policy Planning](https://mlsanity.com/paper/2305.13660/) (2023)
- [Zero-Shot Goal-Directed Dialogue via RL on Imagined Conversations](https://mlsanity.com/paper/2311.05584/) (2023)
- [Q-Probe: A Lightweight Approach to Reward Maximization for Language Models](https://mlsanity.com/paper/2402.14688/) (2024)
- [Planning without Search: Refining Frontier LLMs with Offline Goal-Conditioned RL](https://mlsanity.com/paper/2505.18098/) (2025)
- [On the Effectiveness of Offline RL for Dialogue Response Generation](https://mlsanity.com/paper/2307.12425/) (2023)

## Built on by (most cited first, 8 of 8)

- [Expanding the Capabilities of Reinforcement Learning via Text Feedback](https://mlsanity.com/paper/2602.02482/) (2026)
- [One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL](https://mlsanity.com/paper/2608.12253/) (2026)
- [Asymmetric Actor-Critic for Multi-turn LLM Agents](https://mlsanity.com/paper/2604.00304/) (2026)
- [ALIVE: Awakening LLM Reasoning via Adversarial Learning and Instructive Verbal Evaluation](https://mlsanity.com/paper/2602.05472/) (2026)
- [Listening to the Echo: User-Reaction Aware Policy Optimization via Scalar-Verbal Hybrid Reinforcement Learning](https://mlsanity.com/paper/2603.15434/) (2026)
- [Co-Evolution in Agentic Systems: Toward Self-Directed Evolution Beyond Human Design](https://mlsanity.com/paper/2608.10299/) (2026)
- [CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents](https://mlsanity.com/paper/2608.30147/) (2026)
- [LLM Evolution as an Industry-Scale Ecosystem: A Lifecycle Perspective on Continual Learning](https://mlsanity.com/paper/2606.24901/) (2026)

## Abstract (arXiv)

Large language model (LLM) agents -- LLMs that dynamically interact with an environment over long horizons -- have become an increasingly important area of research, enabling automation in complex tasks involving tool-use, web browsing, and dialogue with people. In the absence of expert demonstrations, training LLM agents has relied on policy gradient methods that optimize LLM policies with respect to an (often sparse) reward function. However, in long-horizon tasks with sparse rewards, learning from trajectory-level rewards can be noisy, leading to training that is unstable and has high sample complexity. Furthermore, policy improvement hinges on discovering better actions through exploration, which can be difficult when actions lie in natural language space. In this paper, we propose Natural Language Actor-Critic (NLAC), a novel actor-critic algorithm that trains LLM policies using a generative LLM critic that produces natural language rather than scalar values. This approach leverages the inherent strengths of LLMs to provide a richer and more actionable training signal; particularly, in tasks with large, open-ended action spaces, natural language explanations for why an action is suboptimal can be immensely useful for LLM policies to reason how to improve their actions, without relying on random exploration. Furthermore, our approach can be trained off-policy without policy gradients, offering a more data-efficient and stable alternative to existing on-policy methods. We present results on a mixture of reasoning, web browsing, and tool-use with dialogue tasks, demonstrating that NLAC shows promise in outperforming existing training approaches and offers a more scalable and stable training paradigm for LLM agents.