Salience Allocation as Guidance for Abstractive Summarization
This work addresses a bottleneck in abstractive summarization for NLP researchers by providing a more adaptable guidance method, though it is incremental as it builds on prior guidance approaches.
The paper tackles the problem of using extractive summaries as guidance for abstractive summarization, which can be overly strict and lead to information loss, by proposing SEASON, a method that uses salience allocation as flexible guidance. Results show it adapts well to articles with varying abstractiveness and achieves effective performance on benchmark datasets, with empirical analysis revealing a natural 15-50 salience split for news article sentences.
Abstractive summarization models typically learn to capture the salient information from scratch implicitly. Recent literature adds extractive summaries as guidance for abstractive summarization models to provide hints of salient content and achieves better performance. However, extractive summaries as guidance could be over strict, leading to information loss or noisy signals. Furthermore, it cannot easily adapt to documents with various abstractiveness. As the number and allocation of salience content pieces vary, it is hard to find a fixed threshold deciding which content should be included in the guidance. In this paper, we propose a novel summarization approach with a flexible and reliable salience guidance, namely SEASON (SaliencE Allocation as Guidance for Abstractive SummarizatiON). SEASON utilizes the allocation of salience expectation to guide abstractive summarization and adapts well to articles in different abstractiveness. Automatic and human evaluations on two benchmark datasets show that the proposed method is effective and reliable. Empirical results on more than one million news articles demonstrate a natural fifteen-fifty salience split for news article sentences, providing a useful insight for composing news articles.