N-gram driven search for sentences in a syntactic network

The text reconstruction method described in this paper finds in a large N-gram-network near-optimal paths that match well with the erroneous input strings to be corrected. The paths in the network and an input string are primarily compared by a redundant hash-addressing method which makes the search very selective and especially fast in spite of holding the network in a disk file during processing. The final selection of a near-optimal path is carried out using the unweighted Levenshtein -distance criterion. The computation-bound dynamic programming algorithm needs to be applied only to selected portions of the strings, however.