Exponential mask local score (ExpLocalScore)
We apply both modifications together: the exponential mask score is log(weighted_coocc) * sigmoid(expPMI - 0.5). Here weighted_coocc is the sum of the weights of the positions of the corpus.