HaRiM$^+$: Evaluating Summary Quality with Hallucination Risk

TL;DR

HaRiM+ is a reference-free, model-based metric for evaluating summary quality by measuring hallucination risk, achieving high correlation with human judgments without additional training.

Contribution

We introduce HaRiM+, a novel hallucination risk metric that requires no extra training and correlates well with human assessments of summary quality.

Findings

HaRiM+ achieves state-of-the-art correlation with human judgments.

It requires only an off-the-shelf summarization model and token likelihoods.

HaRiM+ is reference-free and easy to deploy.

Abstract

One of the challenges of developing a summarization model arises from the difficulty in measuring the factual inconsistency of the generated text. In this study, we reinterpret the decoder overconfidence-regularizing objective suggested in (Miao et al., 2021) as a hallucination risk measurement to better estimate the quality of generated summaries. We propose a reference-free metric, HaRiM+, which only requires an off-the-shelf summarization model to compute the hallucination risk based on token likelihoods. Deploying it requires no additional training of models or ad-hoc modules, which usually need alignment to human judgments. For summary-quality estimation, HaRiM+ records state-of-the-art correlation to human judgment on three summary-quality annotation sets: FRANK, QAGS, and SummEval. We hope that our work, which merits the use of summarization models, facilitates the progress of both automated evaluation and generation of summary.