Proceedings of the Annual Meeting of the Cognitive Science Society

Generalizations that hold across all languages (linguistic universals) provide important insights into cognition, language, and learning. In semantics, the best-known universal is determiner conservativity: the truth of sentences like “every/most/some/no fish swim(s)” depends only on the determiner's first argument (“fish”). This rules out cross-linguistically unattested determiners (e.g., “equi fish swims” meaning ‘the fish and the swimmers are numerically equivalent' isn't conservative because both fish and swimmers matter). Zuber & Keenan (2019) propose a weakening of conservativity: determiners depend on their first OR second argument, but not both. Which constraint do learners obey? We test whether adults are able to learn novel determiners that are classically non-conservative but are conservative on the weakened view. We compare these ‘weakly conservative' cases against novel determiners that are conservative on both views and non-conservative on both views. We find that adults can learn conservative meanings, but not weakly conservative meanings, supporting the classical understanding.