Here is the solution I am using now. It is an adaptation of Peter Norvig's spell checker. It uses word2vec ordering of words to approximate word probabilities.