A method of estimating translation probabilities includes converting a set of word alignment scores into translation probabilities and modifying the translation probabilities based on a bilingual corpus to produce modified translation probabilities. Under many embodiments, the translation probabilities are modified using an iterative training algorithm.