Prove Your Point!: Bringing Proof-Enhancement Principles to Argumentative Essay Generation

TL;DR

This paper presents a unified two-stage framework: Proof-Enhancement and Self-Annotation (PESA) for AEG with a focus on logical enhancement and proposes a tree planning approach that introduces proof principles and ensures logical consistency.

Abstract

Argumentative essay generation (AEG) aims to generate complete texts on specific controversial topics or debates. Although current AEG methods can generate individual opinions, they often overlook the high-level connections between these opinions. This often leads to the generated results being mired in logical confusion, unable to proof their own arguments effectively. The generated essay may present evidence that contradicts the claims or they may fail to assemble the claims into logical flow. In this paper, we present a unified two-stage framework: Proof-Enhancement and Self-Annotation (PESA) for AEG with a focus on logical enhancement. Specifically, we first construct pseudo-labels for logical information,claims and grounds, using a large language model. We then propose a tree planning approach that introduces proof principles and ensures logical consistency. Extensive experimental results show that, benefiting from proof principle guidance, PESA generates argumentative essays with better logical validity and persuasiveness than strong baseline models.

Figures (5)

• Figure 1: Two examples of proof and logical disorganization leading to impaired persuasiveness. Obviously, the upper example gives self-contradiction claim and ground, the lower example gives correct and persuasive proof.
• Figure 2: The full flow chart of PESA. The upper figure shows the Proof-Enhancement process of generating text-planning from writing prompt and finally generating argumentative text, while the lower figure shows the Self-Annotation process of gradually building pseudo-labels for Proof-Enhancement training from ground truth using GPT-4. Detailed Proof-Enhancement samples are given in and Appendix \ref{['logical_structure']} and Appendix \ref{['text-planning-design-app']} .
• Figure 3: PESA compared to other baselines. Human raters compared different model generations and and scored them accordingly.
• Figure 4: Example of the logical structure in human-authored argumentative text. The leftmost writing prompt extends two Major claims, after which each Major claim expands into several grounds or evidence.
• Figure 5: The specific design of Proof-Enhancementg. Two levels of text-planning are shown from top to bottom: the first level is the claim planning contains major claim, and the second level is the ground planning contains grounds, evidence and writing material.