Lecture 3 summary

1
Lecture 12 Summary

• Assuming that the correct word differs from the
  misspelling by a single insertion/deletion/substit
  ution, a list of candidate words is generated
  from the typo (which differ from the misspelling
  by at-most 1 error).
• We then score each correction using
  and choose the
  correct word as the candidate with highest
  probability.
• The prior, P(C) is estimated as
• C count
  of candidate word in the corpus, V number of
  different words in corpus
• Likelihood P(t/c) is estimated by simple
  heurestics governed by the source of typo.
• Confusion Matrix a 26x26 matrix which
  represents of times one letter was incorrectly
  used instead of another (4 confusion matrices
  used for computing likelihoods one for each
  type of error deletion, insertion, substitution,
  reversal).

2

• Lecture 12 Summary

• Minimum Edit Distance
• Previous algorithm assumes that each word has a
  single spelling error.
• To handle multiple errors, we must have a
  technique comparing the distance between
  strings.
• The Minimum Edit Distance between two strings
  is the minimum number of editing operations
  needed to transform one string to the other (can
  be represented as a trace, alignment or an
  operation list) (also known as Levenshtein
  distance ).
• Computed by employing dynamic programming (by
  creating a distance matrix with one column for
  each symbol in the target sequence and one row
  for each symbol in the source sequence).
• Each cell edit_distancei,j contains the
  distance between the first i characters of the
  target and the first j characters of the source.
  The value in each cell is computed as the minimum
  of the 3 possible paths it can follow