Contextualized Rewriting for Text Summarization

Extractive summarization suffers from irrelevance, redundancy and incoherence. Existing work shows that abstractive rewriting for extractive summaries can improve the conciseness readability. These systems consider extracted as only input, which is relatively focused but lose important background knowledge. In this paper, we investigate contextualized rewriting, ingests entire original document. We formalize a seq2seq problem with group alignments, introducing tag solution to model identifying through content-based addressing. Results show our approach significantly outperforms non-contextualized without requiring reinforcement learning, achieving strong improvements on ROUGE scores upon multiple summarizers.