One of the interesting findings from the BioCreative II NER task was that a combination of the outputs of individual systems could achieve notably better performance than any single system (
http://genomebiology.com/2008/9/S2/S2). As U-compare integrates several NER systems, a combination function might have substantial benefits for a moderate implementation effort. Would it be difficult to implement e.g. simple voting-based combination of NER systems?