A General Contextualized Rewriting Framework for Text Summarization

The rewriting method for text summarization combines the advantage of extractive and abstractive approaches, improving conciseness readability summaries. Exiting systems take sentences as only input rewrite each sentence independently, which may lose critical background knowledge break cross-sentence coherence summary. To this end, we propose contextualized to consume entire document maintain summary coherence, representing a part encoding introducing group-tags align We further general framework with an external extractor joint internal extractor, selection special token prediction. demonstrate framework's effectiveness by implementing three rewriter instances on various pre-trained models. Experiments show that significantly outperforms previous non-contextualized rewriting, achieving strong improvements ROUGE scores upon multiple extractors. Empirical results suggest modeling can largely enhance performance.