Horizontal transfers are a primary aspect of languages evolution

More than sixty years ago Morris Swadesh proposed to measure the similarity between two languages by considering their overlap, i.e., the percentage of cognates. The overlap was estimated by comparing two ad hoc lists with words corresponding to same meanings for the two languages. The notable assumption of Swadesh was that replacements in a vocabulary occur at a universal constant rate so that the time distance from the eventual common ancestor can be determined. In his view only replacements are relevant while horizontal transfers (borrowings from other languages) are less important and their effect can be eventually taken into account assuming that the divergence is diminished by contact. Later, the mainstream of glottochronology adopted the point of view that the effect of loanwords could be eliminated by careful work devoted to their identification so that they would not affect any measure of distance between languages. The aim of this paper is to show by experimental evidence that horizontal transfers, on the contrary, are a primary aspect of languages evolution since their effect on the vocabulary is at least as important as that of spontaneous replacements. We finally show that this phenomenon severely and unavoidably limits the possibility to fully reconstruct a proto-language. This limitation is fundamental, i.e., it gives a bound which cannot be infringed, independently of the method used for the reconstruction