Commonsense Evidence Generation and Injection in Reading Comprehension

TL;DR

This paper introduces CEGI, a framework that enhances reading comprehension models by generating and injecting commonsense evidence from language models and knowledge graphs, leading to improved reasoning and accuracy.

Contribution

The paper presents a novel framework combining textual and factual commonsense evidence generation with advanced encoding and capsule networks for better comprehension.

Findings

CEGI outperforms state-of-the-art on CosmosQA with 83.6% accuracy.

Incorporating commonsense evidence improves reasoning in reading comprehension.

The framework effectively combines language models and knowledge graphs for evidence generation.

Abstract

Human tackle reading comprehension not only based on the given context itself but often rely on the commonsense beyond. To empower the machine with commonsense reasoning, in this paper, we propose a Commonsense Evidence Generation and Injection framework in reading comprehension, named CEGI. The framework injects two kinds of auxiliary commonsense evidence into comprehensive reading to equip the machine with the ability of rational thinking. Specifically, we build two evidence generators: the first generator aims to generate textual evidence via a language model; the other generator aims to extract factual evidence (automatically aligned text-triples) from a commonsense knowledge graph after graph completion. Those evidences incorporate contextual commonsense and serve as the additional inputs to the model. Thereafter, we propose a deep contextual encoder to extract semantic relationships among the paragraph, question, option, and evidence. Finally, we employ a capsule network to extract different linguistic units (word and phrase) from the relations, and dynamically predict the optimal option based on the extracted units. Experiments on the CosmosQA dataset demonstrate that the proposed CEGI model outperforms the current state-of-the-art approaches and achieves the accuracy (83.6%) on the leaderboard.