An Improved Algorithm for Learning to Perform Exception-Tolerant Abduction