A computer method is disclosed for ranking word similarities which is applicable to a variety of dictionary applications such as synonym generation, linguistic analysis, document characterization, etc. The method is based upon transforming an input word string into a key word which is invariant for certain types of errors in the input word, such as the doubling of letters, consonant/vowel transpositions, consonant/consonant transpositions. The specific mapping technique is a morphological mapping which generates keys which will have similarities that can be detected during a subsequent ranking procedure. The mapping is defined such that unique consonants of the input word are listed in their original order followed by the unique vowels for the input words, also in their original order. The keys thus generated will be invariant for consonant/vowel transpositions or doubled letters. The utility of the keys is further improved by arranging the consonants in the keys in alphabetical order followed by arranging the vowels in the keys in alphabetical order. The resultant mapping is insensitive to consonant/consonant transpositions, as well as consonant/vowel transpositions and doubled letters. The method then continues by applying a ranking technique which makes use of a compound measure of similarity for ranking the key words. By first measuring the number of basic operations needed to convert an input-derived key word into a dictionary-derived key word (the higher the number, the less similar are the words) and then secondly measuring the length of identical character segments in each pair of key words being matched (the longer the length, the greater the similarity), there is developed a scoring system for ranking the similarity of an input word to dictionary-derived key words, which ignores misspellings in the input word