Preparing data
Now, lists of 40 meanings are accumulated for all languages of the sample – if a word list for a particular language contains more items, they are excluded from further consideration. (However, even these shorter 40-word lists sometimes contain gaps.)
Now, all vowels are treated as a single class; all consonants are collapsed into four classes: bilabials (b, p, f, v); nasals (m, n); fricative velars and uvulars (x, ʁ, etc), the rest of consonants are collapsed into one more class.
For each pair of languages, L’ and L’’, only the meanings present in both lists are kept. Let M denote the number of remaining meanings. For each remaining pair of words  and  , the Levenshtein distance  is computed – disregarding whether or not the two words correspond to a same meaning. The similarity  is then defined in the, following manner:
Share with your friends: |
