# From Heuristic to Analytic: Cognitively Motivated Strategies for Coherent Physical Commonsense Reasoning

**Authors:** Zheyuan Zhang, Shane Storks, Fengyuan Hu, Sungryull Sohn, Moontae Lee, Honglak Lee, Joyce Chai  
**arXiv:** [2310.18364](https://arxiv.org/abs/2310.18364)  
**Date:** 2023-10-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2310.18364v1/

## Key points

- The paper explores limitations of pre-trained language models (PLMs) in generating coherent and reliable reasoning chains.
- The authors propose incorporating interlinked dual processes known as Heuristic-Analytic Reasoning (HAR) to address this challenge.
- The proposed method significantly improves the coherence and reliability of PLM reasoning by adopting human-like reasoning strategies inspired by cognitive psychology.
- State-of-the-art results on a challenging language understanding task demonstrate the effectiveness of the proposed method.
- The findings contribute to advancing our understanding of how cognitive processes can enhance natural language processing models' performance in real-world applications.

## Summary

The paper titled "From Heuristic to Analytic: Cognitively Motivated Strategies for Coherent Physical Commonsense Reasoning" explores the limitations of pre-trained language models (PLMs) in generating coherent and reliable reasoning chains. The authors propose incorporating interlinked dual processes known as Heuristic-Analytic Reasoning (HAR) to address this challenge. By drawing inspiration from cognitive psychology and adopting human-like reasoning strategies, the proposed method significantly improves the coherence and reliability of PLM reasoning. This is demonstrated through state-of-the-art results on a challenging language understanding task. The findings contribute to advancing our understanding of how cognitive processes can enhance natural language processing models' performance in real-world applications.