Abductive Reasoning with Probabilistic Commonsense

2026-05-08

AI Summary

This work addresses a critical limitation in existing neuro-symbolic systems, which rely on formal solvers while implicitly assuming universal agreement in commonsense judgments and neglecting individual variability in commonsense knowledge. To overcome this, the paper introduces PACS, a probabilistic abductive reasoning framework that explicitly models heterogeneity in individual commonsense beliefs. PACS leverages large language models to sample diverse commonsense premises, integrates them with a formal logical solver to generate proofs, and aggregates results to produce population-level probabilistic truth assessments. By rejecting the assumption of uniform consensus, PACS achieves state-of-the-art performance across multiple benchmarks, significantly outperforming chain-of-thought prompting, existing neuro-symbolic approaches, and search-based reasoning methods, thereby enabling scalable and more human-aligned commonsense reasoning.

Abstract

Recent efforts to improve the reasoning abilities of Large Language Models (LLMs) have focused on integrating formal logic solvers within neurosymbolic frameworks. A key challenge is that formal solvers lack commonsense world knowledge, preventing them from making reasoning steps that humans find obvious. Prior methods address this by using LLMs to supply missing commonsense assumptions, but these approaches implicitly assume universal agreement on such commonsense facts. In reality, commonsense beliefs vary across individuals. We propose a probabilistic framework for abductive commonsense reasoning that explicitly models this variation, aiming to determine whether most people would judge a statement as true or false. We introduce Probabilistic Abductive CommonSense (PACS), a novel algorithm that uses an LLM and a formal solver to sample proofs as observations of individuals' distinct commonsense beliefs, and aggregates conclusions across these samples. Empirically, PACS outperforms chain-of-thought reasoning, prior neurosymbolic methods, and search-based approaches across multiple benchmarks.

Problem

• Abductive Reasoning
• Probabilistic Commonsense
• Neurosymbolic Reasoning
• Commonsense Knowledge
• Individual Variation

Innovation

• Probabilistic Abductive Reasoning
• Commonsense Knowledge
• Neurosymbolic Integration
• LLM-Augmented Reasoning
• Belief Variation Modeling