Fuzzy lexical matching

Being able to automatically correct spelling errors is useful in cases where the set of documents is too vast to involve human interaction. In this bachelor's thesis, we investigate an implementation that attempts to perform such corrections using a lexicon and edit distance measure. We compare the familiar Levenshtein and Damerau-Levenshtein distances to modi cations where each edit operation is assigned an individual weight. We nd that the primary bene t of using this form of edit distance over the original is not a higher rate of correction, but a lower susceptibility to false friends. However, deriving the correct weights for each edit operation turns out to be a harder problem than anticipated. While a weighted edit distance can theoretically be implemented e ectively, a deeper analysis of the costs of edit operations is necessary to make such an approach practical.