A Very Fast Associative Method for the Recognition and Correction of Misspelt Words, Based on Redundant Hash Addressing

摘要

A new software method for the recognition and correction of misspelt character strings (words) is introduced in this paper. It was primarily intended to the postprocessing of computer-recognized speech, but might find applications, for instance, in character reading and document retrieval. The basic idea may be applicable to other types of segmented patterns, too. The proposed principle is essentially a hash table method by which, in a multiple search process, statistical match numbers are computed for candidates in a dictionary. By virtue of hash coding, the best candidates are found directly, whereby the speed of this method is greatly superior to any comparison method in which the whole dictionary must be retrieved.