Conclusion
Two properties of multi-document clusters (cluster-based sentence utility and cross-sentence information subsumption) are important for MD summarization
Utility-based single and multiple document evaluation techniques
Centroid-based summarization outperforms lead-based summaries and a combination of the two works even better
Subsumption can be used in the evaluation of multi-document summaries
Offline evaluation of summarizers
Two user studies: on utility and subsumption: interjudge agreement higher in the first study