Efrat-Kowalsky, N.; Ranacher, P.; Neureiter, N.; Widmer, P.; Bickel, B.

It is an unresolved question to what extent the current frequency distributions of linguistic features inform us about what is representative of the language faculty and does not instead result from historical contingencies. We probe this question by leveraging unique data from the oldest attested languages, those preserved through writing from up to 5,000 years ago in the Ancient Near East. We examine 70 grammatical features for which there is sufficient evidence in the available records. After controlling for relatedness we find robust deviations of two of the oldest languages, Hurrian and Sumerian from both the ancient languages and the modern distribution. The spatial and temporal placement of these two languages reveal a divergent distribution of features in the region in prehistory, suggesting massive transformations of the linguistic distributions in the past few millennia. This challenges inferences about general characteristics of language based on modern distributions.