BRIO: Bringing Order to Abstractive Summarization
This addresses performance degradation in abstractive summarization for NLP researchers and practitioners, offering a novel training approach with measurable improvements.
The paper tackles the problem of abstractive summarization models degrading during inference due to deterministic training assumptions, proposing a non-deterministic training paradigm that assigns probability mass based on summary quality, achieving new state-of-the-art results of 47.78 ROUGE-1 on CNN/DailyMail and 49.07 ROUGE-1 on XSum.
Abstractive summarization models are commonly trained using maximum likelihood estimation, which assumes a deterministic (one-point) target distribution in which an ideal model will assign all the probability mass to the reference summary. This assumption may lead to performance degradation during inference, where the model needs to compare several system-generated (candidate) summaries that have deviated from the reference summary. To address this problem, we propose a novel training paradigm which assumes a non-deterministic distribution so that different candidate summaries are assigned probability mass according to their quality. Our method achieves a new state-of-the-art result on the CNN/DailyMail (47.78 ROUGE-1) and XSum (49.07 ROUGE-1) datasets. Further analysis also shows that our model can estimate probabilities of candidate summaries that are more correlated with their level of quality.