A BE-based Multi-document Summarizer with Sentence Compression

This paper describes a multi-document summarizer based on basic elements (BE), head-modifier-relation representation of document content developed at ISI. To increase the coverage of automatically created summaries at a given length, we first generate a summary about twice of the intended length, then apply compression techniques to make sure the resulting summaries fall within the length constraint of target summaries. Our initial results show that the BE-based summarizer with compression achieved 0.0654 in BE-F score that was significantly better than the BE-F score of 0.0542 without compression.