Discourse-Aware Graph Networks for Textual Logical Reasoning

TL;DR

This paper introduces discourse-aware graph networks (DAGNs) that model logical structures within texts to improve question-answering tasks requiring logical reasoning, demonstrating effectiveness and transferability across datasets.

Contribution

The paper proposes a novel logic structural-constraint modeling approach with DAGNs that construct and learn from logical graphs for textual reasoning tasks.

Findings

DAGNs effectively model logical relations in texts.

Logic features improve QA performance.

Features transfer well to unseen logical texts.

Abstract

Textual logical reasoning, especially question-answering (QA) tasks with logical reasoning, requires awareness of particular logical structures. The passage-level logical relations represent entailment or contradiction between propositional units (e.g., a concluding sentence). However, such structures are unexplored as current QA systems focus on entity-based relations. In this work, we propose logic structural-constraint modeling to solve the logical reasoning QA and introduce discourse-aware graph networks (DAGNs). The networks first construct logic graphs leveraging in-line discourse connectives and generic logic theories, then learn logic representations by end-to-end evolving the logic relations with an edge-reasoning mechanism and updating the graph features. This pipeline is applied to a general encoder, whose fundamental features are joined with the high-level logic features for answer prediction. Experiments on three textual logical reasoning datasets demonstrate the reasonability of the logical structures built in DAGNs and the effectiveness of the learned logic features. Moreover, zero-shot transfer results show the features' generality to unseen logical texts.