SummEdits: Measuring LLM Ability at Factual Reasoning Through The Lens of Summarization

Philippe Laban, Wojciech Maciej Kryscinski, Divyansh Agarwal, Alexander Fabbri, Caiming Xiong, Shafiq Joty, Chien-Sheng Wu

Submission Type: Regular Long Paper

Submission Track: Natural Language Generation

Submission Track 2: Summarization

Keywords: factual consistency, faithfulness, summarization, LLMs, benchmark

TL;DR: We create a new benchmark -- SummEdits -- to measure whether LLMs can reason about the consistency of facts in summarization

Abstract:

With the recent appearance of LLMs in practical settings, having methods that can effectively detect factual inconsistencies is crucial to reduce the propagation of misinformation and improve trust in model outputs.
When testing on existing factual consistency benchmarks, we find that a few large language models (LLMs) perform competitively on classification benchmarks for factual inconsistency detection compared to traditional non-LLM methods. However, a closer analysis reveals issues with existing evaluation benchmarks, affecting evaluation precision.
To address this, we propose a new protocol for inconsistency detection benchmark creation and implement it in a 10-domain benchmark called SummEdits. This new benchmark is 20 times more cost-effective per sample than previous benchmarks and highly reproducible, as we estimate inter-annotator agreement at about 0.9.
Most LLMs struggle on SummEdits, with performance close to random chance. The best-performing model, GPT-4, is still 8% below estimated human performance, highlighting the gaps in LLMs' ability to reason about facts and detect inconsistencies when they occur.

Submission Number: 844