Notes
Slide Show
Outline
1
A Probabilistic
Spell Checker
  • Keith Alcock
    • Ling 696f
    • 5 May 2003
2
Overview
  • Noisy channel model
  • Dictionary
    • Sampling, storage
  • Misspeller
    • Algorithm, rules
  • Results
    • Correcting bad words, good words
  • Demo
    • Untrained and trained versions
3
Noisy channel model
4
Dictionary
  • Downsampled from larger versions
5
Dictionary
  • Stored as trie for space and time efficiency
6
Misspeller
  • Based on minimum edit distance algorithm
  • Each arc type corresponds to a spelling rule.
7
Results
  • Correcting bad words
8
Results
  • Correcting good (i.e., real) words
9
Demo