SRLGRN: Semantic Role Labeling Graph Reasoning Network

TL;DR

This paper introduces SRLGRN, a graph reasoning network leveraging semantic role labeling to improve multi-hop question answering by modeling sentence semantics and reasoning paths, enhancing explainability and performance.

Contribution

It presents a novel heterogeneous document-level graph that integrates SRL structures for reasoning, which is a new approach in multi-hop QA tasks.

Findings

Achieves competitive results on HotpotQA benchmark.

Enhances explainability of reasoning paths.

Effectively models cross-paragraph reasoning.

Abstract

This work deals with the challenge of learning and reasoning over multi-hop question answering (QA). We propose a graph reasoning network based on the semantic structure of the sentences to learn cross paragraph reasoning paths and find the supporting facts and the answer jointly. The proposed graph is a heterogeneous document-level graph that contains nodes of type sentence (question, title, and other sentences), and semantic role labeling sub-graphs per sentence that contain arguments as nodes and predicates as edges. Incorporating the argument types, the argument phrases, and the semantics of the edges originated from SRL predicates into the graph encoder helps in finding and also the explainability of the reasoning paths. Our proposed approach shows competitive performance on the HotpotQA distractor setting benchmark compared to the recent state-of-the-art models.