Speeding up inference in Markov logic networks by preprocessing to reduce the size of the resulting grounded network