Error bars for lexicostatistical estimates, with a case study comparing the diversity of Chinese and Romance

This paper applies statistical techniques for measuring sampling error to lexicostatistics, a field in which error has often been discussed, but only rarely measured. We specifically calculate a margin of error for lexicostatistical comparisons based on Swadesh-type vocabulary lists, and use chi-squared tests to estimate a minimum threshold for when two lexicostatistical measurements will be statistically significantly different from one another. The article includes charts which mathematically unsophisticated scholars can easily use to check margins or error. We use margin of error calculations to test the claim that the relative internal diversity of Romance “languages” and Chinese “dialects” is equivalent, finding that no result is possible with extant lexicostatistical studies. We end by suggesting that lexicostatistical dendrograms depict uncertainty with “fat branches,” that is, branches whose width corresponds to statistical uncertainty.