Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metrics

TL;DR

This paper introduces FRANK, a benchmark for factuality metrics in abstractive summarization, highlighting the limitations of existing binary factuality assessments and providing a detailed analysis of factual errors.

Contribution

It proposes a typology of factual errors, creates a benchmark dataset with human annotations, and evaluates the performance of factuality metrics against human judgments.

Findings

Factuality metrics vary in accuracy and specific error detection.

Many metrics do not align well with human judgments.

The typology reveals common error types in summarization models.

Abstract

Modern summarization models generate highly fluent but often factually unreliable outputs. This motivated a surge of metrics attempting to measure the factuality of automatically generated summaries. Due to the lack of common benchmarks, these metrics cannot be compared. Moreover, all these methods treat factuality as a binary concept and fail to provide deeper insights into the kinds of inconsistencies made by different systems. To address these limitations, we devise a typology of factual errors and use it to collect human annotations of generated summaries from state-of-the-art summarization systems for the CNN/DM and XSum datasets. Through these annotations, we identify the proportion of different categories of factual errors in various summarization models and benchmark factuality metrics, showing their correlation with human judgment as well as their specific strengths and weaknesses.