Brief description: Using ngrams to correct mistyped letters in a QWERTY keyboard

Read report.pdf for a more detailed description of the project.

You can run one of the pre-set tests in keyboardCorrection.py by uncommenting the test and running:

$ python keyboardCorrection.py

Note that test4-7 can take a considerably amount of time (>1hr).

Additional tests can also be easily written, as long as the following steps are followed:

1. A model object is created

    model = ngramsModel()
   

2. A confusion matrix is created

    model.generateSTDConfusionMatrix()
   

   or

    model.readConfusionMatrix('customKeyboard.csv')
   

3.1 Either an N-gram model is explicitly created and used

model.createNgramModel(N, corpus)
model.proposeCorrection(WORD, lastKeyOnly = True/False, printOutput=True/False)

3.2 Or CV is performed

model.performCV(corpus, N, k = (>2), lastKeyOnly = True/False)

• Correct multiple errors
• Support more advanced smoothing, backoff, etc
• Evaluate different confusion matrices