Interestingly, the question of whether all languages descend from one is still controversial. Analyses of faraway languages have revealed degrees of similarity, but whether this is statistically significantly different from chance for the most distant branches of the language tree remains controversial. Here I propose a method to unambiguously answer this question.

Take a table whose columns correspond to languages and whose rows correspond to meanings. Each cell contains the word for the corresponding meaning in the corresponding language. If, for every pair of languages, the similarity between vocabularies, using any metric of similarity which assigns shorter distances for vocabularies that are more similar or easier to evolve from one another, the average distance between the languages is statistically significantly smaller than that between fictitious vocabularies where the rows have been rearranged randomly for each column, then all languages can be said to be related. Furthermore, the same can be said if this can be done for reconstructed ancestors corresponding to all branchpoints in an evolutionary tree of languages. Indeed, as long as the statistical test properly accounts for multiple comparisons, all that is needed is that each language be related to the rest via a connected graph (a graph such that there exists a path between all pairs of vertices). If we are to assay the origin of language, we must ensure that any similarity due to later lateral influences between languages are ignored. Concentrating on words for primitive concepts, whose words tend to be old and change little, is one way to do this.

 Create your own free wiki!Help Add to Del.ici.ous Add to Digg Add to Reddit Add to fUrl Add to SpurlPublic, not yet premium Privacy Policy RSS