An improved statistical test for historical linguistics

Historical linguistics needs procedures to evaluate the similarity between languages through the comparison of specific word lists drawn from the whole vocabulary. The main issue is to evaluate a fair threshold for the number of similar items beyond which it is sensible to reject the hypothesis of chance similarity. After a short review of papers dealing with that problem, in this paper an extension of those methods is proposed which exploits available data in a more efficient way. In particular, the exact distribution of the new test statistics is calculated and the power of the new procedure is compared with the power of the existing method.