Effect of Learning Argument Reconstruction on LLM Critical Thinking

Determine whether large language models can enhance their critical thinking ability by learning to reconstruct arguments, where argument reconstruction denotes extracting both explicit and implicit premises and structuring their logical connections into a premise–conclusion form that supports evaluation.

Background

The paper adopts a three-stage operationalization of critical thinking—argument identification, argument reconstruction, and argument evaluation—and emphasizes argument reconstruction as central because it makes inferential commitments explicit and enables subsequent evaluation.

Prior work on argument reconstruction with LLMs largely focuses on short, deductive, fallacy-free, and domain-specific arguments, limiting generalizability. This motivates the question of whether training LLMs to perform argument reconstruction can improve their broader critical thinking capabilities.