Pointwise Mutual Information Calculator

Pointwise mutual information measures how much more often two events occur together than they would if they were independent. Enter the joint and marginal probabilities, or the raw counts and the size of the corpus they came from, and pick a logarithm base. The result gives PMI, positive PMI, normalised PMI, the observed-to-expected ratio and the largest PMI the rarer of the two events allows. It is the standard association score for word collocations, and like every count-based score it is unstable when the joint count is very small.