Argument Reconstruction as Supervision for Critical Thinking in LLMs

TL;DR

This paper introduces GAAR, an automatic argument reconstruction engine, and Arguinas, a high-quality dataset, demonstrating that training models on argument reconstruction improves critical thinking tasks in LLMs.

Contribution

The paper presents a novel automatic argument reconstruction engine and a new dataset, showing that learning argument reconstruction enhances LLMs' critical thinking abilities.

Findings

Models trained on Arguinas outperform others in critical thinking tasks.

Training on argument reconstruction yields significant performance gains.

The GAAR engine effectively reconstructs diverse arguments automatically.

Abstract

To think critically about arguments, human learners are trained to identify, reconstruct, and evaluate arguments. Argument reconstruction is especially important because it makes an argument's underlying inferences explicit. However, it remains unclear whether LLMs can similarly enhance their critical thinking ability by learning to reconstruct arguments. To address this question, we introduce a holistic framework with three contributions. We (1) propose an engine that automatically reconstructs arbitrary arguments (GAAR), (2) synthesize a new high-quality argument reconstruction dataset (Arguinas) using the GAAR engine, and (3) investigate whether learning argument reconstruction benefits downstream critical thinking tasks. Our experimental results show that, across seven critical thinking tasks, models trained to learn argument reconstruction outperform models that do not, with the largest performance gains observed when training on the proposed Arguinas dataset.