Fact-driven Logical Reasoning for Machine Reading Comprehension

TL;DR

This paper introduces a hierarchical fact-based formalism for logical reasoning in machine reading comprehension, effectively integrating both commonsense and temporary knowledge to improve reasoning capabilities.

Contribution

It proposes a novel formalism of knowledge units and a supergraph model that captures sentence-level and entity-level interactions, enhancing reasoning performance.

Findings

Significant improvements on logical reasoning benchmarks

Effective integration of commonsense and temporary facts

Model generalizes across different backbone architectures

Abstract

Recent years have witnessed an increasing interest in training machines with reasoning ability, which deeply relies on accurately and clearly presented clue forms. The clues are usually modeled as entity-aware knowledge in existing studies. However, those entity-aware clues are primarily focused on commonsense, making them insufficient for tasks that require knowledge of temporary facts or events, particularly in logical reasoning for reading comprehension. To address this challenge, we are motivated to cover both commonsense and temporary knowledge clues hierarchically. Specifically, we propose a general formalism of knowledge units by extracting backbone constituents of the sentence, such as the subject-verb-object formed ``facts''. We then construct a supergraph on top of the fact units, allowing for the benefit of sentence-level (relations among fact groups) and entity-level interactions (concepts or actions inside a fact). Experimental results on logical reasoning benchmarks and dialogue modeling datasets show that our approach improves the baselines substantially, and it is general across backbone models. Code is available at \url{https://github.com/ozyyshr/FocalReasoner}.