Debate-to-Write: A Persona-Driven Multi-Agent Framework for Diverse Argument Generation

TL;DR

The study tackles the difficulty of generating persuasive, viewpoint-rich arguments by introducing a persona-based multi-agent framework that simulates human debate to plan long-form arguments. The method partitions the task into persona assignment, debate-driven text planning, and separate argument writing, all via prompting without additional training. Empirical results show improved perspective diversity and competitive persuasiveness against strong baselines, validated by both automatic metrics and human judgments. This approach advances controllable, diverse argument generation with potential applications in education, policy analysis, and automated discourse. The framework also highlights the importance of explicit planning and structured collaboration among models for coherent, multi-perspective writing.

Abstract

Writing persuasive arguments is a challenging task for both humans and machines. It entails incorporating high-level beliefs from various perspectives on the topic, along with deliberate reasoning and planning to construct a coherent narrative. Current language models often generate surface tokens autoregressively, lacking explicit integration of these underlying controls, resulting in limited output diversity and coherence. In this work, we propose a persona-based multi-agent framework for argument writing. Inspired by the human debate, we first assign each agent a persona representing its high-level beliefs from a unique perspective, and then design an agent interaction process so that the agents can collaboratively debate and discuss the idea to form an overall plan for argument writing. Such debate process enables fluid and nonlinear development of ideas. We evaluate our framework on argumentative essay writing. The results show that our framework can generate more diverse and persuasive arguments through both automatic and human evaluations.

Figures (11)

• Figure 1: The overview of our framework. Given an input topic, our framework first assigns distinct personas to each agent, each representing a unique perspective relevant to the topic. The agents then engage in discussions and debates to refine their ideas and develop a high-level plan. Finally, an argument writing module transforms this plan into a surface argumentative essay. A complete sample output with intermediate results for each step is shown in Figure \ref{['fig:additonal_samples2']}.
• Figure 2: Snippet of the debate among agents for example in Figure \ref{['fig:overall']}. The right structure shows the logical flow, where solid arrow is oppose relation and dashed arrow is support.
• Figure 3: Different Agent persona assignment for the same topic.
• Figure 4: The interface of human evaluations.
• Figure 5: The full generated argument plan and output for input in Figure \ref{['fig:overall']} and Figure \ref{['fig:sample_discussion']}