Computational linguistics

Why might two text-processing applications tokenize the same contraction differently?