spellcorrect

This package implements a 1990 paper A Spelling Correction Program Based on a Noisy Channel Model by Mark D. Kernighan, Kenneth Ward Church and William A. Gale. A copy of this paper can be found online here or through a simple search.

A package to correct non-word spelling error in sentences using ngram MAP Language Models, Noisy Channel Model, Error Confusion Matrix and Damerau-Levenshtein Edit Distance.

I have encoded the datasets in a computer readable-form. Feel free to utilize the dataset and the ngram helper library (included) to improve this library or make your own. Or you can play with the hyper-parameters to improve your results. You can ping me on #IRC and Twitter at @jbhoosreddy if you have any questions.

class SpellCorrect
 |  A program to correct non-word spelling error in sentences using ngram MAP Language Models, Noisy Channel Model, Error Confusion Matrix and Damerau-Levenshtein Edit Distance.
 |  Usage:
 |  Input: 'she is a briliant acress'
 |  Response: she is a brilliant actress
 |  
 |  Methods defined here:
 |  
 |  __init__(self)
 |      Constructor method to load external nGram class, load words, confusion matrix and dictionary.
 |  
 |  channelModel(self, x, y, edit)
 |      Method to calculate channel model probability for errors.
 |  
 |  dlEditDistance(self, s1, s2)
 |      Method to calculate Damerau-Levenshtein Edit Distance for two strings.
 |  
 |  editType(self, candidate, word)
 |      Method to calculate edit type for single edit errors.
 |  
 |  genCandidates(self, word)
 |      Method to generate set of candidates for a given word using Damerau-Levenshtein Edit Distance.
 |  
 |  loadConfusionMatrix(self)
 |      Method to load Confusion Matrix from external data file.
 |  
 |  loadDict(self)
 |      Method to load dictionary from external data file.

Confusion matrices

As answered by @lzw429 in #1, the authors have provided in the appendix the computed deletion, addition, substitution and reversal matrices. I have manually transcribed this data to .data files which are processed in memory. The paper has more information on how this data was originally created by the authors.