# Summary Refinement through Denoising

## TL;DR

This paper introduces a post-processing method that uses text-to-text rewriting models trained on synthetic noise to improve summary quality by reducing redundancy and correcting out-of-context information.

## Contribution

It presents a novel approach for refining summaries through denoising models trained on synthetic noise, enhancing existing summarization systems.

## Key findings

- Improved metric scores on summarization benchmarks.
- Reduced redundancy in generated summaries.
- Effective correction of out-of-context information.

## Abstract

We propose a simple method for post-processing the outputs of a text summarization system in order to refine its overall quality. Our approach is to train text-to-text rewriting models to correct information redundancy errors that may arise during summarization. We train on synthetically generated noisy summaries, testing three different types of noise that introduce out-of-context information within each summary. When applied on top of extractive and abstractive summarization baselines, our summary denoising models yield metric improvements while reducing redundancy.