A model and a fast algorithm for multiple errors spelling correction

Typographical Nearest-Neighbor Search in a Finite-State Lexicon and Its Application to Spelling Correction

“…The first approach is more often admitted (e.g. in [4,14] since the right correction candidate for a misspelled word may not be its nearest neighbour. In our opinion, the second approach is preferable for many applications for three reasons: statistical studies show that words with multiple errors are rare (0.17% till 1.99% of unknown words in a corpus, with [15], users are easily discouraged by long lists of correction candidates, and the search time grows exponentially with the admitted distance threshold.…”

“…The distance measure between two strings admitted in this paper, as well as in Oflazer's one, is the error distance of Du and Chang [4] (although Oflazer still uses the notion of edit distance), as described in the previous section. There is no theoretical distance limit between an erroneous word and its right correction.…”

“…In [4] this distance measure is modified and renamed to error distance by assigning cost 1 to each editing operation and by admitting that errors occur in linear order from left to right so that a later operation may not cancel the effect of an earlier operation. Thus, inversions occur only between letters that are adjacent in the original word and remain adjacent in the erroneous word (e.g.…”

Robust Spelling Correction

“…For the latter, techniques look for possible editing sequences to reflect the error occurrence phenomenon in spelling. These strategies study correction patterns, most of them taking into account the edit distance [5], but also on occasion introducing constraints on the spelling process [6] in order to cut down the computational time needed for the correction.…”

Automatic Spelling Correction in Galician

“…So, Oflazer [5] introduces a cut-off distance that can be performed efficiently by maintaining a matrix [2] which help the system to determine when a partial repair will not yield any result by providing non-decreasing repair paths. In order to save this maintaining, Savary [6] embeds the distance in the repair algorithm, although this allows to partial corrections may be reached several times with different intermediate distances; which is not time-efficient for error threshold values bigger than one.…”