Short, frequent words are more likely to appear genetically related by chance

Abstract

An important question in historical linguistics is whether deep genetic relationships exist across language families. Although specific families can be reconstructed back to around 6,000 y ago, Pagel et al. (1) claim that seven Eurasian families arose from a common ancestor 15,000 y ago. Pagel et al. develop a phylogenetic model, starting with a subset of the Swadesh basic word list for seven language families in the Languages of the World Etymological Database, which lists reconstructed proto-words and cognates. Because these reconstructions are potentially unreliable, Pagel et al. treat each reconstructed cognate pair as a binary random variable. They find a robust correlation between the size of the cognate class and the word replacement rate (i.e., how fast the word is likely to be replaced in the vocabulary), which is closely related to frequency. As predicted, words with a slower replacement rate show deeper relationships across language families, which they take as evidence that there are deep relationships among the seven families.