To improve stochastic phonographic transduction based on a set of letter-phoneme correspondences, a sample of incorrect machine pronunciations was given to a linguist who is also an English language teacher. He found that errors frequently arose with silent-e markings and affixation. This paper examines the effect on performance when silent-e and/or affix correspondences are explicitly added to the existing set. Results show that adding silent-e correspondences did not significantly improve performance, but adding affix correspondences did when testing with both the training set and unseen words. However, there was no significant improvement for proper names and novel words.