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.
@article{arxiv.1907.10873,
title = {Summary Refinement through Denoising},
author = {Nikola I. Nikolov and Alessandro Calmanovici and Richard H. R. Hahnloser},
journal= {arXiv preprint arXiv:1907.10873},
year = {2019}
}