Ranking Candidates

The second stage scores each correction. Let t be the typo and c range over a set C of candidate corrections. The most likely correction is then

c = argmax

The prior probability P(c) can be estimated by

• Counting how often the word c appears in the corpus
• Normalising the count by dividing it by the number N of words in the corpus. Zero counts can cause problems, and so we add .5 to all counts (this is called ``smoothing''). Having done this, we must compensate by adding 0.5*V to the denominator for each word V in the vocabulary so that

  P(c) =