Summary Generation and Evaluation in SumUM

We describe and evaluate Sum UM, a text summarization system that produces indicative-informative abstracts for technical papers. Our approach consists of the shallow syntactic and conceptual analysis of the source document and of the implementation of text regeneration techniques based on a study of abstracts produced by professional abstractors. In an evaluation of indicative content in a categorization task, we observed no differences with other automatic method, while differences are observed in an evaluation of informative content. In an evaluation of text quality, the abstracts were considered acceptable when compared with other automatic abstracts.