receiving an input search query;

identifying a set of candidate word sequences;

a processor determining a distributional similarity between a word of the input search query and a term of one of the candidate word sequences using a query log of logged search queries by:

identifying a set of all co-occurrence words that co-occur with the word of the input search query in at least one logged search query in the query log and that also co-occur with the term of one of the candidate word sequences in at least one logged search query in the query log;

for each co-occurrence word in the set of identified co-occurrence words:

determining the number of logged search queries in which the co-occurrence word and the word of the input search query appeared together in the query log to form a first co-occurrence frequency; and

determining the number of logged search queries in which the co-occurrence word and the term of the candidate word sequence appeared together in the query log to form a second co-occurrence frequency;

using the first and second co-occurrence frequencies for each co-occurrence word in the set of all co-occurrence words to determine the distributional similarity using a metric from a set of metrics consisting of a confusion probability metric and a cosine metric, wherein the confusion probability metric is calculated by taking a sum over all co-occurrence words in the set of all co-occurrence words where each summand in the sum is computed based at least on a product of a probability of the word of the input search query given the co-occurrence word and a probability of the term of the candidate word sequence given the co-occurrence word and wherein the cosine metric is calculated by determining the cosine of an angle between a first vector for the word of the input search query and a second vector for the term of the candidate word sequence;

using the distributional similarity to determine an error model probability that describes the probability of the input search query given the candidate word sequence associated with the distributional similarity;

using the error model probability to determine a probability of the candidate word sequence associated with the distributional similarity given the input search query; and

using the probability of the candidate word sequence associated with the distributional similarity given the input search query to select a candidate word sequence as a corrected word sequence for the search query.