Augmenting training sets with synthetic examples for authorship verification may not consistently improve classifier performance.
