18
Discussion
lCalibration of
relatedness with similarity data
lCalibration
point inaccurate
lSubstitution
errors untested
lSemantic bias
in human typing errors not addressed
lBinary
threshold not best choice
lFrequency on
synset, word, or word sense