Nothing fancy, just n-grams (on the character level). Did some normalization by converting likelihoods to quantiles, though.