Abstract
Languages encode similar semantic features in their kinship terminologies, the set of words denoting family members, but different languages seem to prioritise different distinctions. Is any configuration of kinship semantics possible, or are there commonalities across languages in which distinctions are encoded? We use computational methods to identify clear cross-linguistic tendencies in how six semantic features are encoded in kinship terms. Across 1242 languages, all encode information on Generation, while other features like Gender or Seniority are less consistently encoded. Importantly, the features are encoded in a predictable order of importance. We find that this tendency mirrors participants' behaviour in a linguistic categorisation experiment, suggesting kinship semantics reflect cognitive biases active at the level of the individual. Additionally, despite claims that kin terms are a straightforward reflection of social practice, we find varying evidence for this; kin term semantics systematically correspond with patterns of descent, but not marriage norms.