As a suggestion, you could search the line for the substring with the least Levenshtein distance from the correction, to use as the text to replace—and thus also support the "*word" correction syntax I see quite more often from non-technical friends. :)