ABSUM: a Knowledge-Based Abstractive Summarizer

ive summarization is one of the main goals of text summarization research, but also one of its greatest challenges. The authors of a recent literature review (Lloret and Palomar 2012) even conclude that “abstractive paradigms [...] will become one of the main challenges to solve” in text summarization. In building an abstractive summarization system, however, it is often hard to imagine where to begin and how to proceed in order to incorporate some kind of semantic understanding of the source documents to create a shorter text that contains only the relevant elements for the task at hand. This paper introduces the Knowledge-Based Abstractive Summarization (K-BABS) approach, to address various summarization tasks and domains in a flexible and scalable way. Its architecture relies on an analysis of the source documents and on a task blueprint. This resource describes how to transform the representation of the text into natural language for the summary. It implicitly encodes knowledge about the summarization task into rules applied by the summarization system. The task blueprint, which can be constructed automatically, semiautomatically, or manually, guides every step of the summarization process. © 2005 Association for Computational Linguistics Computational Linguistics Volume xx, Number xx Annotated Parse Trees