next up previous
Next: The problem

A spelling corrector incorporating knowledge of English orthography and pronunciation

Brett Kessler

Abstract:

I have developed a spelling corrector, correct, based on a model treating misspellings as flawed attempts to spell words using the sound-spelling correspondences of English orthography. On a common test suite, it generates a correct guess for 97% of the misspellings. It ranks multiple guesses by the Levenshtein distance from the misspelling, combined with the frequency of the sound-spelling correspondences presumed to have generated the misspelling. The average rank of the correct respelling, using a complete English dictionary as the universe of candidate respellings, is 1.1. This paper describes the development and structure of the correct programme.





  • LaTeX source file.
  • Gzipped PostScript file.

  • Up to Brett Kessler's home page.

    Webster: Brett Kessler
    [email protected]
    Last change Feb. 16, 1999