Correcting Diverse Factual Errors in Abstractive Summarization via Post-Editing and Language Model Infilling
This addresses the issue of unreliable summaries for users of abstractive summarization systems, though it is incremental as it builds on existing post-editing approaches.
The paper tackled the problem of factual errors in abstractive summarization by proposing a post-editing model trained on synthetic non-factual summaries generated via language model infilling, resulting in improvements of over ~11 points on CNN/DM and ~31 points on XSum in factuality scores.
Abstractive summarization models often generate inconsistent summaries containing factual errors or hallucinated content. Recent works focus on correcting factual errors in generated summaries via post-editing. Such correction models are trained using adversarial non-factual summaries constructed using heuristic rules for injecting errors. However, generating non-factual summaries using heuristics often does not generalize well to actual model errors. In this work, we propose to generate hard, representative synthetic examples of non-factual summaries through infilling language models. With this data, we train a more robust fact-correction model to post-edit the summaries to improve factual consistency. Through quantitative and qualitative experiments on two popular summarization datasets -- CNN/DM and XSum -- we show that our approach vastly outperforms prior methods in correcting erroneous summaries. Our model -- FactEdit -- improves factuality scores by over ~11 points on CNN/DM and over ~31 points on XSum on average across multiple summarization models, producing more factual summaries while maintaining competitive summarization quality.