Proceedings of the Annual Meeting of the Cognitive Science Society

Grammatical number is one of the most widespread morphological categories in the world's languages. When only a single number distinction is expressed, it is almost always between singular and non-singular (plural), and even in systems with additional values such as dual, non-singular categories are frequently collapsed. The collapse of singular and dual categories is less frequent but attested, whereas groupings that merge singular with plural are virtually unattested. These patterns reveal strong cross-linguistic asymmetries in how number systems are partitioned. Linguistic theories have proposed that such regularities reflect underlying feature primitives, such as singular and minimal (Harbour, 2014; Noyer, 1992; Silverstein, 1976). The former explains why dual and plural constitute a natural non-singularity category, and the latter explains why singular and dual do so, as both are the cardinally smallest singular and non-singular values. However, little is known about whether these formal features correspond to learners' cognitive representations. In this paper, we test this proposal using two artificial language learning experiments that manipulate alternative number partitions. Our results show that learners systematically prefer partitions consistent with the proposed feature structure, with strongest preferences for singular-based contrasts, followed by minimal-based distinctions. These findings provide initial evidence for a cognitively principled explanation of the typology of partitions of grammatical number systems.