Statistical Determination of Linguistic Relationship

The Mathematical Assessment of Long‐Range Linguistic Relationships

“…Several researchers have addressed the question of how many CVC (consonant–vowel–consonant) matches one could expect to find when comparing words (Swadesh 1954; Cowan 1962; Bender 1969; Ringe 1999). The answer depends on how similar sounds must be in order to be considered a match, and how many languages are being compared.…”

“…In the bilateral case, Swadesh (1954) used probability theory to demonstrate that only about four matches would be expected per hundred words. Using different formulas, Cowan (1962) concluded that no more than three matches would turn up in 95% of chance cases. Bender (1969) took the empirical tack of counting the number of matches between pairs of 21 unrelated languages and lowered the number to two per hundred.…”

“…It is not at all uncommon for researchers to model what a morpheme can look like in a language by defining it as a string of sounds, each different sound having an equal chance of occurring in each position of the morpheme. Some approximations get better, or add slop factors (Swadesh 1954; Cowan 1962), but rarely does one see a model that takes into account the real inventories, phoneme frequencies and phonotactics of the languages under consideration, not to mention morpheme level constraints: whether the language is likely to repeat the same consonant in a root, how likely it is to tolerate two roots that are homophonous, and so forth.…”