Learning First-Order Probabilistic Models with Combining Rules