Computing

Kernighan et al. ignored most of these factors and simply counted the occurrences of particular kinds of error occurring in a large corpus of errors.

Using this technique they constructed a series of confusion matrices for the different kinds of error. For example a substitution confusion matrix would be a 26x26 table

• sub(x,y) - number of times x is substituted for y.
• ins(x,y) - number of times x was typed as xy
• del(x,y) - number of times xy was typed as x
• tran(x,y) - number of times xy was typed as yx