A small tool for generating transposition, deletion, and substitution errors.
For every one of these operations, it is possible to come up with a multitude of alternatives for a given word.
For example, the word hospital can be transformed into hopsital and hpsoital with a single transposition. It is clear that the former obeys the orthotactic rules of english more closely than the latter one.
To pick between these alternatives, all operations are returned in rank order, where the rank is decided according to the absolute difference between the mean bigram frequency of the original word and the edited word.
Additional constraints are that a returned candidate can not be a word in the reference corpus, i.e., all generated words are non-words, and that the returned candidate can not be equal to the input word, i.e., even if the input word is a non-word, the generated word can not be the same word.
Note that outer letters are never transposted, deleted, or substituted. This is because there is a wealth of psycholinguistic evidence that changing these letters changes the nature of the processing with regard to transposed, deleted, and substituted letters.
from trnsps import Trnsps
my_corpus = ["godless", "sodless", "respondent", "stockless", "miles"]
t = Trnsps(my_corpus)
result = list(t.transposition(my_corpus))MIT