Errors in sentence can be of two types: Individual word errors and errors in a sentence. An aritle related to this work written by me can be found here
An error at word level can be identified if the word does not exist in dictionary. A metric like Edit Distance can be used to find nearest words from the dictionary to the misspelt word. Another metric like Weighted Edit Distance can also be used.
If all the words in a sentece are present in dictionary but don't match the context, it is a real-word error. All possible candidates (words at edit distance 1 from given word) are considered. Using a bigram language model, the probabilities of candidate senetences is calculated.
