Improving the Readability of Scientific Concept Analogies with Cognitive Conflict Reinforcement Learning
Year: 2024
Type: article
Abstract: Large language models are increasingly being used for education and science communication by automatically generating explanations of scientific concepts. However, prior research has found that the analogies produced by LLMs lack human-like psycholinguistic properties important for readability. In this work, we propose cognitive conflict reinforcement learning (CCRL) to improve the psycholinguistic properties of analogies generated by LLMs. Specifically, we create cognitive conflict between the original LLM and a cloned LLM during reinforcement learning. This helps address the cognitive rigidity problem in LLMs. Experimental results demonstrate that our approach significantly outperforms existing RL algorithms and human performance in improving various readability metrics of generated analogies. Authors: Yuang Cai: Beijing University of Posts and Telecommunications; Yuyu Yuan: Beijing University of Posts and Telecommunications; Jinsheng Shi: Beijing University of Posts and Telecommunications; Rui Han: Beijing University of Posts and Telecommunications; Zhenyu Zhao: Beijing University of Post and Telecommunications ; Zijie Shi: Beijing University of Posts and Telecommunications
DOI: 10.48448/hnar-hz84
Authors: Cognitive Science Society 2024, Cai, Yuang
Institution: Beijing University of Posts and Telecommunications
Language: English
Subfield: Linguistics and Language
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W6948161718