Argument Mining (AM) is fundamentally constrained by the scarcity of high-quality structure-annotated datasets. While LLMs have shown promise in synthetic data generation, producing synthetic AM data that is both structurally accurate and sufficiently diverse remains a challenging problem. To address this problem, we revisit synthetic data generation for AM from a new perspective and propose a novel Adversarial Reinforcement Learning framework for Data Synthesis. The proposed framework jointly optimizes the generator and the discriminator in an adversarial loop, in which the generator produces structured AM instances, and the discriminator provides learning signals by distinguishing real data from synthetic candidates. This enables the generator to progressively improve both the structural accuracy of generated argument data while maintaining diversity through adversarial feedback. Extensive experiments demonstrate that the proposed framework consistently improves AM performance on three benchmark datasets in both full-data and low-resource settings, validating its effectiveness and scalability.

Argumentative essay generation (AEG) aims to produce coherent and persuasive essays on controversial claims. Most existing methods follow a plan-then-write paradigm, where a single plan is generated and used to guide long-form writing. However, the quality of the generated essay heavily depends on the initial plan, and errors or omissions in planning often propagate to the final output. To address this limitation, we propose the Plan–Write–Refine framework, an iterative approach for AEG. It first generates an initial plan and a draft essay, then abstracts the essay to obtain condensed feedback, and refines the plan accordingly. Compared with the prior static planning methods, our framework enables iterative plan improvement and better structural organization of the essay by closing the loop between planning and writing. We conduct extensive experiments on four AEG benchmarks. Both automatic and human evaluations show that our framework achieves consistent improvements over strong baselines.