Spelling Alteration for Web

Search phrase correction is the challenging problem of proposing alternative versions of search queries typed into a web search engine. Any number of different approaches can be taken to solve this problem, each having different strengths. Recently, the availability of large datasets that include web corpora have increased the interest in purely data-driven spellers. In this paper, a hybrid approach that uses traditional dictionary-based spelling tools in combination with data-driven probabilistic techniques is described. The performance of this approach was sufficient to win Microsoft’s Speller Challenge in June, 2011.