Argument Reconstruction as Supervision for Critical Thinking in LLMs
This addresses the challenge of improving LLMs' reasoning abilities for applications requiring critical thinking, though it is incremental as it builds on existing methods for argument analysis.
The paper tackles the problem of enhancing critical thinking in large language models (LLMs) by training them to reconstruct arguments, and finds that models trained on a new synthesized dataset (Arguinas) outperform those that do not across seven critical thinking tasks.
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. The source code and dataset will be publicly available.