I was curious because there is a much larger training corpus for Lisps, so if the problem really is one of training data rather ordering then this would be a way of showing that.