An advanced Spell Checker using the Russian Language based on the search log of one the most popular Russian search engine Mail.Ru. The system uses Machine Learning and Nautral Language Processing techniques.

The spell checker can successfully resolve 4 types of mistypes:

• Layout errors
• Grammar errors
• Incorrect split
• Incorrect join

The system consists of:

• Language model, P(queue) (n-gramm probability model of the requests)
• Error model, P(fix|orig) - the probability that user sent orig-request but actually wanted to send fix-request. Uses weighted Levenshtein distance.
• Fixes generator based on fuzzy string search to improve efficiency
• Machine Learning mistype classifier
• Iterative fixes (for mixed type of mistype)