On the other hand Schultz and Pedersen found 7% to 14% improvement on a baseline when they added a sense disambiguation module though they did not use a MRD as a reference source for senses. K&C and Sanderson and most others have used an MRD to define the senses of words as reference points but for S&P, their assumption was that the senses of a polysemous word are defined by clusters of common contexts surrounding the word in question, not from a static, predefined, list of possibly incomplete or alternativey too fine, senses.