Faithful or Extractive? On Mitigating the Faithfulness-Abstractiveness Trade-off in Abstractive Summarization

• DOI: 10.18653/v1/2022.acl-long.100
• Publication Date: 2022-06-03T01:34:53Z
• Authors: Faisal Ladhak; Esin Durmus; He He; Claire Cardie; Kathleen McKeown
• Keywords: Baseline (sea)

Despite recent progress in abstractive summarization, systems still suffer from faithfulness errors. While prior work has proposed models that improve faithfulness, it is unclear whether the improvement comes an increased level of extractiveness model outputs as one naive way to make summarization more extractive. In this work, we present a framework for evaluating effective systems, by generating faithfulness-abstractiveness trade-off curve serves control at different operating points on abstractiveness spectrum. We then show Maximum Likelihood Estimation (MLE) baseline well recently methods improving fail consistently over same abstractiveness. Finally, learn selector identify most faithful and summary given document, system can attain higher scores human evaluations while being than two datasets. Moreover, our able achieve better