Inasmuch as the theory behind ` correct` is that misspellings are
just perturbations in the normal process of generating spellings by
applying the rules of English sound-spelling correspondences, one
might hypothesize that misspellings are more likely to be fairly
similar to the correct word than widely different. For if there is a
certain probability that a person would generate the wrong spelling
correspondence for any particular phoneme in the word, then the
probability that several or all should be wrong should be smaller.
Indeed, [Pollock 1984] reports that in keycoded text, 90-95% of all
misspellings have but one error. Therefore some measure of the
distance between a misspelling and a candidate respelling would seem
to be one good way of ranking among several candidate respellings.

The module ` Kruskal` calculates the Levenshtein distance between
two words. It follows the algorithm set forth by Kruskal in
[Sankoff 1983], under the simple set of assumptions that all insertions,
deletions, and substitutions are of equal weight. Then it returns
an integer number telling the minimum number of such operations that
would be needed to transform the one word into the other.

Wed Dec 27 22:16:48 PST 1995