Discourse-Driven Evaluation: Unveiling Factual Inconsistency in Long Document Summarization

TL;DR

This paper introduces a discourse-based framework to detect factual inconsistencies in long document summaries, improving evaluation accuracy by leveraging discourse features and sentence decomposition.

Contribution

It proposes a novel discourse-inspired method that decomposes texts and uses discourse information to enhance factual inconsistency detection in long summaries.

Findings

Improved performance over baseline models on multiple benchmarks.

Factual errors are more common in complex sentences with certain discourse features.

Incorporating discourse features significantly enhances factual inconsistency detection.

Abstract

Detecting factual inconsistency for long document summarization remains challenging, given the complex structure of the source article and long summary length. In this work, we study factual inconsistency errors and connect them with a line of discourse analysis. We find that errors are more common in complex sentences and are associated with several discourse features. We propose a framework that decomposes long texts into discourse-inspired chunks and utilizes discourse information to better aggregate sentence-level scores predicted by natural language inference models. Our approach shows improved performance on top of different model baselines over several evaluation benchmarks, covering rich domains of texts, focusing on long document summarization. This underscores the significance of incorporating discourse features in developing models for scoring summaries for long document factual inconsistency.