Concretized Proposition Prompting Resolves Composition-Knowledge Dichotomy in Large Language Models

Researchers propose Concretized Proposition Prompting (CPP) to resolve the composition-knowledge dichotomy in large language models, enhancing reasoning performance especially in medical benchmarks. CPP bridges composition- and knowledge-based approaches, providing logically organized and factually grounded reasoning across various foundation models.

arXiv:2607.08018v1 Announce Type: new Abstract: LLMs often struggle to balance compositionality with knowledgeability, a challenge we define as Composition-Knowledge Dichotomy. To address this, we propose Concretized Proposition Prompting CPP , a framework that explicitly concretizes propositions relevant to questions. The results demonstrate that CPP significantly enhances reasoning performance, particularly in medical benchmarks where precise knowledge is paramount, while being competitive on math benchmarks where deductive reasoning is prioritized. Additional experiments reveal that CPP is scalable to various foundation models and parameter sizes, being a fundamental paradigm that bridges the gap between composition- and knowledge-based approaches. Consequently, CPP resolves the composition-knowledge dichotomy by providing a solid foundation for logically organized and factually grounded reasoning.