Entity-based SpanCopy for Abstractive Summarization to Improve the Factual Consistency

TL;DR

This paper introduces an entity-based SpanCopy mechanism to enhance factual consistency in abstractive summarization by reducing entity mismatches, demonstrating improved accuracy without sacrificing saliency.

Contribution

The paper proposes a novel entity-based SpanCopy method and its extension with Global Relevance to improve factual consistency in summarization models.

Findings

Significant improvement in entity-level factual accuracy

No notable change in saliency metrics

Effective across multiple datasets

Abstract

Despite the success of recent abstractive summarizers on automatic evaluation metrics, the generated summaries still present factual inconsistencies with the source document. In this paper, we focus on entity-level factual inconsistency, i.e. reducing the mismatched entities between the generated summaries and the source documents. We therefore propose a novel entity-based SpanCopy mechanism, and explore its extension with a Global Relevance component. Experiment results on four summarization datasets show that SpanCopy can effectively improve the entity-level factual consistency with essentially no change in the word-level and entity-level saliency. The code is available at https://github.com/Wendy-Xiao/Entity-based-SpanCopy