SummExecEdit: A Factual Consistency Benchmark in Summarization with Executable Edits

TL;DR

SummExecEdit introduces a new benchmark using executable edits to evaluate models on factual consistency detection and explanation in summarization, revealing current models' significant challenges and common explanation errors.

Contribution

The paper presents SummExecEdit, a novel benchmark with executable edits for assessing factual error detection and explanation in summarization models, enhancing interpretability and challenge.

Findings

Top model scores 0.49 joint detection and explanation

Over 30% of errors in 20+ LLMs

45.4% explanation errors focus on unrelated summary parts

Abstract

Detecting factual inconsistencies in summarization is critical, yet existing benchmarks lack the necessary challenge and interpretability for robust evaluation. In this paper, we introduce SummExecEdit, a novel pipeline and benchmark leveraging executable edits to assess models on their ability to both detect factual errors and provide accurate explanations. The top-performing model, Claude3-Opus, achieves a joint detection and explanation score of only 0.49 in our benchmark, with individual scores of 0.67 for detection and 0.73 for explanation. We conduct detailed evaluations to assess the current state of models in this field and find that more than half of the 20+ LLMs in our study struggle with over 30% of the SummExecEdit benchmark. Additionally, we identify four primary types of explanation errors, with 45.4% of them involving a focus on completely unrelated parts of the summary.